training.log

dev_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/entail-val.hdf5
embedding_size 300
test_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/entail-test.hdf5
weight_decay 1e-05
dev_interval 1
train_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/entail-train.hdf5
max_grad_norm 5
gpu_id 0
para_init 0.01
lr 0.05
log_fname log41.log
epoch 250
max_length 9999
w2v_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/glove.hdf5
log_dir /disk/scratch/bowenli/nmt/struct-attn/data/snli/experiment/
Adagrad_init 0.0
optimizer Adagrad
hidden_size 300
display_interval 1000
model_path /disk/scratch/bowenli/nmt/struct-attn/data/snli/experiment/
loading training data...
train size # sent 549366
dev size # sent 9841
test size # sent 9823
loading input embeddings...
start to train...
dev_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/entail-val.hdf5
embedding_size 300
test_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/entail-test.hdf5
weight_decay 1e-05
dev_interval 1
train_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/entail-train.hdf5
max_grad_norm 5
gpu_id 0
para_init 0.01
lr 0.05
log_fname log41.log
epoch 250
max_length 9999
w2v_file /disk/scratch/bowenli/nmt/struct-attn/data/snli/baseline/glove.hdf5
log_dir /disk/scratch/bowenli/nmt/struct-attn/data/snli/experiment/
Adagrad_init 0.0
optimizer Adagrad
hidden_size 300
display_interval 1000
model_path /disk/scratch/bowenli/nmt/struct-attn/data/snli/experiment/
loading training data...
train size # sent 549366
dev size # sent 9841
test size # sent 9823
loading input embeddings...
start to train...
epoch 0, batches 1000|18183, train-acc 0.393, loss 3.450, para-norm 2544.106, grad-norm 0.242, time 7.48s, 
epoch 0, batches 2000|18183, train-acc 0.474, loss 1.024, para-norm 2913.896, grad-norm 38.054, time 7.21s, 
epoch 0, batches 3000|18183, train-acc 0.495, loss 0.999, para-norm 2998.982, grad-norm 1.196, time 7.29s, 
epoch 0, batches 4000|18183, train-acc 0.509, loss 0.989, para-norm 2982.814, grad-norm 6.281, time 7.28s, 
epoch 0, batches 5000|18183, train-acc 0.554, loss 0.950, para-norm 3116.135, grad-norm 2.933, time 7.20s, 
epoch 0, batches 6000|18183, train-acc 0.574, loss 0.915, para-norm 3150.346, grad-norm 2.342, time 7.20s, 
epoch 0, batches 7000|18183, train-acc 0.578, loss 0.905, para-norm 3118.841, grad-norm 3.211, time 7.19s, 
epoch 0, batches 8000|18183, train-acc 0.590, loss 0.886, para-norm 3147.054, grad-norm 1.443, time 7.18s, 
epoch 0, batches 9000|18183, train-acc 0.596, loss 0.880, para-norm 3131.193, grad-norm 1.252, time 7.19s, 
epoch 0, batches 10000|18183, train-acc 0.597, loss 0.871, para-norm 3143.333, grad-norm 0.608, time 7.17s, 
epoch 0, batches 11000|18183, train-acc 0.605, loss 0.865, para-norm 3189.787, grad-norm 1.645, time 6.98s, 
epoch 0, batches 12000|18183, train-acc 0.604, loss 0.864, para-norm 3256.163, grad-norm 2.037, time 6.99s, 
epoch 0, batches 13000|18183, train-acc 0.610, loss 0.849, para-norm 3301.233, grad-norm 4.269, time 7.00s, 
epoch 0, batches 14000|18183, train-acc 0.611, loss 0.849, para-norm 3300.800, grad-norm 0.628, time 7.00s, 
epoch 0, batches 15000|18183, train-acc 0.611, loss 0.855, para-norm 3326.044, grad-norm 3.908, time 7.08s, 
epoch 0, batches 16000|18183, train-acc 0.620, loss 0.841, para-norm 3360.388, grad-norm 2.468, time 6.99s, 
epoch 0, batches 17000|18183, train-acc 0.619, loss 0.840, para-norm 3377.955, grad-norm 0.730, time 7.08s, 
epoch 0, batches 18000|18183, train-acc 0.627, loss 0.828, para-norm 3382.633, grad-norm 1.044, time 7.05s, 
epoch 0, batches 18183|18183, train-acc 0.618, loss 0.836, para-norm 3390.012, grad-norm 1.327, time 1.28s, 
dev-acc 0.651
current best-dev:
	0 0.651
save model!
epoch 1, batches 1000|18183, train-acc 0.629, loss 0.827, para-norm 3410.844, grad-norm 4.754, time 7.18s, 
epoch 1, batches 2000|18183, train-acc 0.636, loss 0.822, para-norm 3425.057, grad-norm 0.943, time 6.98s, 
epoch 1, batches 3000|18183, train-acc 0.631, loss 0.821, para-norm 3457.790, grad-norm 1.160, time 6.99s, 
epoch 1, batches 4000|18183, train-acc 0.632, loss 0.817, para-norm 3479.428, grad-norm 2.638, time 6.99s, 
epoch 1, batches 5000|18183, train-acc 0.638, loss 0.813, para-norm 3503.350, grad-norm 0.800, time 6.98s, 
epoch 1, batches 6000|18183, train-acc 0.637, loss 0.810, para-norm 3510.361, grad-norm 0.935, time 6.99s, 
epoch 1, batches 7000|18183, train-acc 0.646, loss 0.806, para-norm 3552.350, grad-norm 2.671, time 7.00s, 
epoch 1, batches 8000|18183, train-acc 0.636, loss 0.815, para-norm 3564.160, grad-norm 1.891, time 7.05s, 
epoch 1, batches 9000|18183, train-acc 0.642, loss 0.805, para-norm 3585.154, grad-norm 1.332, time 6.97s, 
epoch 1, batches 10000|18183, train-acc 0.641, loss 0.807, para-norm 3600.447, grad-norm 4.418, time 6.98s, 
epoch 1, batches 11000|18183, train-acc 0.644, loss 0.799, para-norm 3622.351, grad-norm 1.036, time 6.99s, 
epoch 1, batches 12000|18183, train-acc 0.645, loss 0.798, para-norm 3639.696, grad-norm 1.599, time 6.98s, 
epoch 1, batches 13000|18183, train-acc 0.647, loss 0.796, para-norm 3651.166, grad-norm 1.598, time 6.99s, 
epoch 1, batches 14000|18183, train-acc 0.650, loss 0.794, para-norm 3785.320, grad-norm 2.768, time 6.98s, 
epoch 1, batches 15000|18183, train-acc 0.647, loss 0.800, para-norm 3904.979, grad-norm 0.961, time 6.99s, 
epoch 1, batches 16000|18183, train-acc 0.650, loss 0.794, para-norm 3954.878, grad-norm 2.346, time 6.99s, 
epoch 1, batches 17000|18183, train-acc 0.655, loss 0.786, para-norm 3966.809, grad-norm 1.601, time 6.98s, 
epoch 1, batches 18000|18183, train-acc 0.654, loss 0.786, para-norm 3958.688, grad-norm 2.298, time 6.99s, 
epoch 1, batches 18183|18183, train-acc 0.666, loss 0.779, para-norm 3955.321, grad-norm 1.659, time 1.28s, 
dev-acc 0.682
current best-dev:
	0 0.651
	1 0.682
save model!
epoch 2, batches 1000|18183, train-acc 0.661, loss 0.773, para-norm 3968.602, grad-norm 2.075, time 7.16s, 
epoch 2, batches 2000|18183, train-acc 0.655, loss 0.780, para-norm 3977.764, grad-norm 0.814, time 6.98s, 
epoch 2, batches 3000|18183, train-acc 0.662, loss 0.773, para-norm 3980.521, grad-norm 1356.941, time 6.98s, 
epoch 2, batches 4000|18183, train-acc 0.663, loss 0.775, para-norm 3982.930, grad-norm 0.823, time 6.98s, 
epoch 2, batches 5000|18183, train-acc 0.658, loss 0.775, para-norm 3995.447, grad-norm 2.944, time 7.00s, 
epoch 2, batches 6000|18183, train-acc 0.664, loss 0.773, para-norm 3990.421, grad-norm 2.175, time 6.99s, 
epoch 2, batches 7000|18183, train-acc 0.665, loss 0.765, para-norm 4002.923, grad-norm 2.330, time 6.99s, 
epoch 2, batches 8000|18183, train-acc 0.663, loss 0.771, para-norm 3995.286, grad-norm 2.446, time 6.99s, 
epoch 2, batches 9000|18183, train-acc 0.667, loss 0.768, para-norm 3996.105, grad-norm 15.539, time 6.98s, 
epoch 2, batches 10000|18183, train-acc 0.663, loss 0.774, para-norm 3994.159, grad-norm 2.219, time 7.00s, 
epoch 2, batches 11000|18183, train-acc 0.662, loss 0.770, para-norm 3997.630, grad-norm 2.882, time 6.99s, 
epoch 2, batches 12000|18183, train-acc 0.667, loss 0.761, para-norm 4000.290, grad-norm 4.301, time 6.96s, 
epoch 2, batches 13000|18183, train-acc 0.667, loss 0.766, para-norm 4008.372, grad-norm 1.469, time 6.96s, 
epoch 2, batches 14000|18183, train-acc 0.667, loss 0.759, para-norm 4019.294, grad-norm 26.309, time 6.96s, 
epoch 2, batches 15000|18183, train-acc 0.667, loss 0.762, para-norm 4035.862, grad-norm 3.412, time 6.93s, 
epoch 2, batches 16000|18183, train-acc 0.672, loss 0.757, para-norm 4049.960, grad-norm 7.211, time 6.95s, 
epoch 2, batches 17000|18183, train-acc 0.670, loss 0.759, para-norm 4066.305, grad-norm 1.306, time 6.95s, 
epoch 2, batches 18000|18183, train-acc 0.668, loss 0.763, para-norm 4072.221, grad-norm 176.188, time 6.94s, 
epoch 2, batches 18183|18183, train-acc 0.668, loss 0.760, para-norm 4072.285, grad-norm 1.998, time 1.27s, 
dev-acc 0.700
current best-dev:
	0 0.651
	1 0.682
	2 0.700
save model!
epoch 3, batches 1000|18183, train-acc 0.670, loss 0.764, para-norm 4066.881, grad-norm 1.400, time 7.52s, 
epoch 3, batches 2000|18183, train-acc 0.672, loss 0.754, para-norm 4068.634, grad-norm 1.404, time 6.94s, 
epoch 3, batches 3000|18183, train-acc 0.678, loss 0.745, para-norm 4063.755, grad-norm 2.200, time 6.95s, 
epoch 3, batches 4000|18183, train-acc 0.674, loss 0.754, para-norm 4067.086, grad-norm 1.536, time 6.95s, 
epoch 3, batches 5000|18183, train-acc 0.677, loss 0.748, para-norm 4072.720, grad-norm 1.886, time 7.02s, 
epoch 3, batches 6000|18183, train-acc 0.678, loss 0.744, para-norm 4076.697, grad-norm 2.713, time 6.95s, 
epoch 3, batches 7000|18183, train-acc 0.676, loss 0.749, para-norm 4086.039, grad-norm 3.318, time 6.96s, 
epoch 3, batches 8000|18183, train-acc 0.675, loss 0.748, para-norm 4092.376, grad-norm 3.048, time 7.03s, 
epoch 3, batches 9000|18183, train-acc 0.680, loss 0.740, para-norm 4101.130, grad-norm 438.540, time 6.95s, 
epoch 3, batches 10000|18183, train-acc 0.680, loss 0.744, para-norm 4110.872, grad-norm 5.214, time 6.96s, 
epoch 3, batches 11000|18183, train-acc 0.679, loss 0.742, para-norm 4124.231, grad-norm 1.993, time 6.97s, 
epoch 3, batches 12000|18183, train-acc 0.677, loss 0.741, para-norm 4146.497, grad-norm 13.010, time 6.97s, 
epoch 3, batches 13000|18183, train-acc 0.688, loss 0.731, para-norm 4155.532, grad-norm 1.768, time 6.94s, 
epoch 3, batches 14000|18183, train-acc 0.676, loss 0.751, para-norm 4165.138, grad-norm 3.488, time 6.97s, 
epoch 3, batches 15000|18183, train-acc 0.682, loss 0.739, para-norm 4185.427, grad-norm 1.168, time 6.96s, 
epoch 3, batches 16000|18183, train-acc 0.680, loss 0.741, para-norm 4201.041, grad-norm 3.898, time 6.97s, 
epoch 3, batches 17000|18183, train-acc 0.688, loss 0.730, para-norm 4201.283, grad-norm 2.224, time 7.04s, 
epoch 3, batches 18000|18183, train-acc 0.685, loss 0.733, para-norm 4200.451, grad-norm 1.685, time 6.95s, 
epoch 3, batches 18183|18183, train-acc 0.690, loss 0.726, para-norm 4203.325, grad-norm 5.555, time 1.27s, 
dev-acc 0.717
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
save model!
epoch 4, batches 1000|18183, train-acc 0.683, loss 0.735, para-norm 4213.046, grad-norm 0.616, time 7.08s, 
epoch 4, batches 2000|18183, train-acc 0.692, loss 0.722, para-norm 4233.591, grad-norm 2.506, time 6.96s, 
epoch 4, batches 3000|18183, train-acc 0.687, loss 0.727, para-norm 4233.019, grad-norm 1.245, time 6.94s, 
epoch 4, batches 4000|18183, train-acc 0.684, loss 0.736, para-norm 4224.730, grad-norm 1.280, time 7.02s, 
epoch 4, batches 5000|18183, train-acc 0.688, loss 0.730, para-norm 4232.016, grad-norm 1.137, time 6.94s, 
epoch 4, batches 6000|18183, train-acc 0.689, loss 0.730, para-norm 4249.731, grad-norm 3.400, time 6.94s, 
epoch 4, batches 7000|18183, train-acc 0.686, loss 0.731, para-norm 4243.468, grad-norm 1.015, time 7.01s, 
epoch 4, batches 8000|18183, train-acc 0.692, loss 0.721, para-norm 4259.634, grad-norm 1.474, time 6.94s, 
epoch 4, batches 9000|18183, train-acc 0.689, loss 0.725, para-norm 4275.833, grad-norm 1.777, time 6.95s, 
epoch 4, batches 10000|18183, train-acc 0.691, loss 0.720, para-norm 4278.030, grad-norm 6.725, time 6.96s, 
epoch 4, batches 11000|18183, train-acc 0.689, loss 0.726, para-norm 4287.074, grad-norm 1.097, time 6.95s, 
epoch 4, batches 12000|18183, train-acc 0.688, loss 0.726, para-norm 4297.560, grad-norm 1.917, time 6.96s, 
epoch 4, batches 13000|18183, train-acc 0.688, loss 0.728, para-norm 4294.505, grad-norm 3.199, time 6.95s, 
epoch 4, batches 14000|18183, train-acc 0.687, loss 0.724, para-norm 4285.145, grad-norm 2.429, time 6.95s, 
epoch 4, batches 15000|18183, train-acc 0.692, loss 0.720, para-norm 4285.018, grad-norm 6.513, time 6.95s, 
epoch 4, batches 16000|18183, train-acc 0.690, loss 0.724, para-norm 4283.484, grad-norm 1.686, time 6.92s, 
epoch 4, batches 17000|18183, train-acc 0.689, loss 0.724, para-norm 4289.143, grad-norm 1.672, time 6.94s, 
epoch 4, batches 18000|18183, train-acc 0.697, loss 0.717, para-norm 4290.837, grad-norm 1.801, time 6.93s, 
epoch 4, batches 18183|18183, train-acc 0.695, loss 0.720, para-norm 4288.198, grad-norm 2.103, time 1.27s, 
dev-acc 0.720
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
save model!
epoch 5, batches 1000|18183, train-acc 0.691, loss 0.719, para-norm 4300.103, grad-norm 1.668, time 7.04s, 
epoch 5, batches 2000|18183, train-acc 0.696, loss 0.715, para-norm 4301.055, grad-norm 6.651, time 6.94s, 
epoch 5, batches 3000|18183, train-acc 0.693, loss 0.717, para-norm 4293.733, grad-norm 5.594, time 6.94s, 
epoch 5, batches 4000|18183, train-acc 0.692, loss 0.718, para-norm 4304.692, grad-norm 3.207, time 6.93s, 
epoch 5, batches 5000|18183, train-acc 0.698, loss 0.711, para-norm 4301.722, grad-norm 1.226, time 6.95s, 
epoch 5, batches 6000|18183, train-acc 0.698, loss 0.712, para-norm 4312.368, grad-norm 0.994, time 6.95s, 
epoch 5, batches 7000|18183, train-acc 0.697, loss 0.711, para-norm 4336.321, grad-norm 1.703, time 6.96s, 
epoch 5, batches 8000|18183, train-acc 0.695, loss 0.714, para-norm 4352.389, grad-norm 1.494, time 6.95s, 
epoch 5, batches 9000|18183, train-acc 0.699, loss 0.708, para-norm 4362.373, grad-norm 1.174, time 6.93s, 
epoch 5, batches 10000|18183, train-acc 0.691, loss 0.723, para-norm 4365.996, grad-norm 3.079, time 6.93s, 
epoch 5, batches 11000|18183, train-acc 0.698, loss 0.708, para-norm 4369.777, grad-norm 263.210, time 6.95s, 
epoch 5, batches 12000|18183, train-acc 0.697, loss 0.707, para-norm 4369.800, grad-norm 2.977, time 6.93s, 
epoch 5, batches 13000|18183, train-acc 0.695, loss 0.712, para-norm 4366.320, grad-norm 1.702, time 6.95s, 
epoch 5, batches 14000|18183, train-acc 0.699, loss 0.706, para-norm 4367.124, grad-norm 1.864, time 6.95s, 
epoch 5, batches 15000|18183, train-acc 0.701, loss 0.705, para-norm 4368.065, grad-norm 1.450, time 6.94s, 
epoch 5, batches 16000|18183, train-acc 0.701, loss 0.708, para-norm 4367.799, grad-norm 1.950, time 6.94s, 
epoch 5, batches 17000|18183, train-acc 0.701, loss 0.705, para-norm 4365.316, grad-norm 4.523, time 6.94s, 
epoch 5, batches 18000|18183, train-acc 0.700, loss 0.703, para-norm 4367.874, grad-norm 5.166, time 6.94s, 
epoch 5, batches 18183|18183, train-acc 0.690, loss 0.714, para-norm 4367.441, grad-norm 1.872, time 1.27s, 
dev-acc 0.731
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
save model!
epoch 6, batches 1000|18183, train-acc 0.696, loss 0.709, para-norm 4368.023, grad-norm 1.682, time 7.62s, 
epoch 6, batches 2000|18183, train-acc 0.707, loss 0.700, para-norm 4372.123, grad-norm 1.100, time 7.31s, 
epoch 6, batches 3000|18183, train-acc 0.705, loss 0.696, para-norm 4374.959, grad-norm 5.195, time 6.94s, 
epoch 6, batches 4000|18183, train-acc 0.701, loss 0.704, para-norm 4378.822, grad-norm 2.569, time 6.95s, 
epoch 6, batches 5000|18183, train-acc 0.703, loss 0.697, para-norm 4385.770, grad-norm 2.857, time 6.94s, 
epoch 6, batches 6000|18183, train-acc 0.704, loss 0.696, para-norm 4387.781, grad-norm 2.014, time 6.96s, 
epoch 6, batches 7000|18183, train-acc 0.702, loss 0.700, para-norm 4381.870, grad-norm 2.556, time 6.95s, 
epoch 6, batches 8000|18183, train-acc 0.698, loss 0.701, para-norm 4388.824, grad-norm 5.511, time 6.96s, 
epoch 6, batches 9000|18183, train-acc 0.706, loss 0.699, para-norm 4394.418, grad-norm 2.058, time 6.95s, 
epoch 6, batches 10000|18183, train-acc 0.708, loss 0.694, para-norm 4392.794, grad-norm 2.835, time 6.96s, 
epoch 6, batches 11000|18183, train-acc 0.703, loss 0.695, para-norm 4387.749, grad-norm 1.942, time 6.95s, 
epoch 6, batches 12000|18183, train-acc 0.710, loss 0.692, para-norm 4387.606, grad-norm 1.564, time 6.95s, 
epoch 6, batches 13000|18183, train-acc 0.705, loss 0.696, para-norm 4393.890, grad-norm 1.360, time 6.96s, 
epoch 6, batches 14000|18183, train-acc 0.703, loss 0.700, para-norm 4397.425, grad-norm 5.004, time 6.96s, 
epoch 6, batches 15000|18183, train-acc 0.704, loss 0.697, para-norm 4393.277, grad-norm 1.776, time 6.96s, 
epoch 6, batches 16000|18183, train-acc 0.704, loss 0.697, para-norm 4391.961, grad-norm 31.862, time 6.95s, 
epoch 6, batches 17000|18183, train-acc 0.708, loss 0.689, para-norm 4388.312, grad-norm 0.849, time 6.95s, 
epoch 6, batches 18000|18183, train-acc 0.713, loss 0.685, para-norm 4391.537, grad-norm 1.489, time 6.95s, 
epoch 6, batches 18183|18183, train-acc 0.702, loss 0.693, para-norm 4390.889, grad-norm 4.433, time 1.27s, 
dev-acc 0.738
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
save model!
epoch 7, batches 1000|18183, train-acc 0.709, loss 0.688, para-norm 4393.076, grad-norm 2.116, time 7.02s, 
epoch 7, batches 2000|18183, train-acc 0.706, loss 0.693, para-norm 4400.352, grad-norm 2.656, time 6.95s, 
epoch 7, batches 3000|18183, train-acc 0.712, loss 0.678, para-norm 4404.976, grad-norm 1.839, time 6.95s, 
epoch 7, batches 4000|18183, train-acc 0.713, loss 0.684, para-norm 4407.257, grad-norm 1.821, time 6.96s, 
epoch 7, batches 5000|18183, train-acc 0.717, loss 0.677, para-norm 4410.881, grad-norm 1.619, time 6.94s, 
epoch 7, batches 6000|18183, train-acc 0.717, loss 0.674, para-norm 4413.097, grad-norm 3.012, time 6.93s, 
epoch 7, batches 7000|18183, train-acc 0.714, loss 0.677, para-norm 4415.161, grad-norm 2.675, time 6.94s, 
epoch 7, batches 8000|18183, train-acc 0.715, loss 0.675, para-norm 4417.129, grad-norm 42.177, time 6.93s, 
epoch 7, batches 9000|18183, train-acc 0.712, loss 0.683, para-norm 4415.938, grad-norm 2.502, time 6.93s, 
epoch 7, batches 10000|18183, train-acc 0.711, loss 0.684, para-norm 4415.047, grad-norm 2.375, time 6.93s, 
epoch 7, batches 11000|18183, train-acc 0.716, loss 0.677, para-norm 4415.950, grad-norm 1.614, time 6.92s, 
epoch 7, batches 12000|18183, train-acc 0.711, loss 0.680, para-norm 4422.952, grad-norm 4.295, time 6.93s, 
epoch 7, batches 13000|18183, train-acc 0.712, loss 0.683, para-norm 4423.904, grad-norm 1.288, time 6.93s, 
epoch 7, batches 14000|18183, train-acc 0.711, loss 0.683, para-norm 4421.473, grad-norm 1.254, time 6.94s, 
epoch 7, batches 15000|18183, train-acc 0.712, loss 0.677, para-norm 4425.015, grad-norm 1.404, time 6.94s, 
epoch 7, batches 16000|18183, train-acc 0.716, loss 0.671, para-norm 4424.226, grad-norm 2.244, time 6.93s, 
epoch 7, batches 17000|18183, train-acc 0.715, loss 0.677, para-norm 4424.386, grad-norm 1.121, time 6.93s, 
epoch 7, batches 18000|18183, train-acc 0.716, loss 0.675, para-norm 4426.039, grad-norm 2.166, time 6.93s, 
epoch 7, batches 18183|18183, train-acc 0.731, loss 0.664, para-norm 4426.408, grad-norm 1.095, time 1.27s, 
dev-acc 0.747
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
save model!
epoch 8, batches 1000|18183, train-acc 0.714, loss 0.679, para-norm 4427.104, grad-norm 1.621, time 6.99s, 
epoch 8, batches 2000|18183, train-acc 0.721, loss 0.665, para-norm 4431.391, grad-norm 1.722, time 6.94s, 
epoch 8, batches 3000|18183, train-acc 0.718, loss 0.671, para-norm 4431.812, grad-norm 1.341, time 6.94s, 
epoch 8, batches 4000|18183, train-acc 0.719, loss 0.673, para-norm 4431.372, grad-norm 0.981, time 6.93s, 
epoch 8, batches 5000|18183, train-acc 0.721, loss 0.667, para-norm 4437.956, grad-norm 1.221, time 6.93s, 
epoch 8, batches 6000|18183, train-acc 0.723, loss 0.665, para-norm 4434.484, grad-norm 3.555, time 6.94s, 
epoch 8, batches 7000|18183, train-acc 0.719, loss 0.670, para-norm 4436.167, grad-norm 2.588, time 6.92s, 
epoch 8, batches 8000|18183, train-acc 0.720, loss 0.664, para-norm 4445.838, grad-norm 1.917, time 6.92s, 
epoch 8, batches 9000|18183, train-acc 0.715, loss 0.672, para-norm 4443.121, grad-norm 1.460, time 6.93s, 
epoch 8, batches 10000|18183, train-acc 0.724, loss 0.665, para-norm 4447.154, grad-norm 3.344, time 6.93s, 
epoch 8, batches 11000|18183, train-acc 0.721, loss 0.666, para-norm 4449.651, grad-norm 257.826, time 6.94s, 
epoch 8, batches 12000|18183, train-acc 0.720, loss 0.669, para-norm 4459.639, grad-norm 176.224, time 6.93s, 
epoch 8, batches 13000|18183, train-acc 0.723, loss 0.663, para-norm 4465.565, grad-norm 10.851, time 6.91s, 
epoch 8, batches 14000|18183, train-acc 0.727, loss 0.654, para-norm 4466.144, grad-norm 3.187, time 6.92s, 
epoch 8, batches 15000|18183, train-acc 0.720, loss 0.663, para-norm 4477.106, grad-norm 1.527, time 6.94s, 
epoch 8, batches 16000|18183, train-acc 0.721, loss 0.668, para-norm 4480.862, grad-norm 1.514, time 6.93s, 
epoch 8, batches 17000|18183, train-acc 0.718, loss 0.666, para-norm 4484.935, grad-norm 34.154, time 6.92s, 
epoch 8, batches 18000|18183, train-acc 0.726, loss 0.658, para-norm 4483.661, grad-norm 1.853, time 6.93s, 
epoch 8, batches 18183|18183, train-acc 0.715, loss 0.678, para-norm 4484.691, grad-norm 14.252, time 1.27s, 
dev-acc 0.755
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
save model!
epoch 9, batches 1000|18183, train-acc 0.727, loss 0.654, para-norm 4487.876, grad-norm 2.149, time 7.11s, 
epoch 9, batches 2000|18183, train-acc 0.731, loss 0.654, para-norm 4486.244, grad-norm 1.560, time 6.99s, 
epoch 9, batches 3000|18183, train-acc 0.728, loss 0.652, para-norm 4488.826, grad-norm 1.616, time 6.92s, 
epoch 9, batches 4000|18183, train-acc 0.725, loss 0.658, para-norm 4494.135, grad-norm 2.875, time 6.92s, 
epoch 9, batches 5000|18183, train-acc 0.724, loss 0.659, para-norm 4492.387, grad-norm 1.822, time 6.91s, 
epoch 9, batches 6000|18183, train-acc 0.725, loss 0.654, para-norm 4494.570, grad-norm 1.494, time 6.93s, 
epoch 9, batches 7000|18183, train-acc 0.731, loss 0.651, para-norm 4494.188, grad-norm 1.612, time 6.93s, 
epoch 9, batches 8000|18183, train-acc 0.731, loss 0.646, para-norm 4503.184, grad-norm 2.900, time 6.92s, 
epoch 9, batches 9000|18183, train-acc 0.733, loss 0.646, para-norm 4502.818, grad-norm 1.286, time 6.92s, 
epoch 9, batches 10000|18183, train-acc 0.727, loss 0.653, para-norm 4506.878, grad-norm 1.331, time 6.92s, 
epoch 9, batches 11000|18183, train-acc 0.730, loss 0.652, para-norm 4514.434, grad-norm 4.360, time 6.93s, 
epoch 9, batches 12000|18183, train-acc 0.725, loss 0.657, para-norm 4517.541, grad-norm 326.656, time 6.95s, 
epoch 9, batches 13000|18183, train-acc 0.729, loss 0.652, para-norm 4524.629, grad-norm 64.801, time 6.92s, 
epoch 9, batches 14000|18183, train-acc 0.731, loss 0.649, para-norm 4530.447, grad-norm 2.801, time 6.98s, 
epoch 9, batches 15000|18183, train-acc 0.727, loss 0.654, para-norm 4526.930, grad-norm 2.257, time 6.92s, 
epoch 9, batches 16000|18183, train-acc 0.727, loss 0.650, para-norm 4526.836, grad-norm 1.775, time 6.92s, 
epoch 9, batches 17000|18183, train-acc 0.728, loss 0.649, para-norm 4526.369, grad-norm 3.423, time 6.93s, 
epoch 9, batches 18000|18183, train-acc 0.732, loss 0.652, para-norm 4531.153, grad-norm 2.612, time 6.92s, 
epoch 9, batches 18183|18183, train-acc 0.727, loss 0.655, para-norm 4530.155, grad-norm 54.798, time 1.27s, 
dev-acc 0.757
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
save model!
epoch 10, batches 1000|18183, train-acc 0.731, loss 0.646, para-norm 4532.862, grad-norm 2.492, time 6.99s, 
epoch 10, batches 2000|18183, train-acc 0.729, loss 0.652, para-norm 4533.496, grad-norm 3.658, time 6.92s, 
epoch 10, batches 3000|18183, train-acc 0.734, loss 0.641, para-norm 4532.678, grad-norm 3.334, time 7.21s, 
epoch 10, batches 4000|18183, train-acc 0.735, loss 0.642, para-norm 4535.483, grad-norm 4.311, time 7.17s, 
epoch 10, batches 5000|18183, train-acc 0.733, loss 0.642, para-norm 4539.951, grad-norm 1.322, time 7.18s, 
epoch 10, batches 6000|18183, train-acc 0.732, loss 0.643, para-norm 4540.103, grad-norm 3.035, time 7.17s, 
epoch 10, batches 7000|18183, train-acc 0.731, loss 0.650, para-norm 4537.717, grad-norm 1.370, time 7.17s, 
epoch 10, batches 8000|18183, train-acc 0.733, loss 0.641, para-norm 4541.542, grad-norm 4.505, time 7.16s, 
epoch 10, batches 9000|18183, train-acc 0.735, loss 0.641, para-norm 4542.958, grad-norm 1.532, time 7.18s, 
epoch 10, batches 10000|18183, train-acc 0.726, loss 0.647, para-norm 4543.488, grad-norm 44.975, time 7.18s, 
epoch 10, batches 11000|18183, train-acc 0.739, loss 0.635, para-norm 4542.501, grad-norm 0.781, time 7.23s, 
epoch 10, batches 12000|18183, train-acc 0.734, loss 0.640, para-norm 4538.953, grad-norm 0.977, time 7.22s, 
epoch 10, batches 13000|18183, train-acc 0.732, loss 0.639, para-norm 4541.552, grad-norm 1.880, time 7.24s, 
epoch 10, batches 14000|18183, train-acc 0.736, loss 0.636, para-norm 4542.576, grad-norm 3.437, time 7.21s, 
epoch 10, batches 15000|18183, train-acc 0.732, loss 0.646, para-norm 4540.244, grad-norm 1.326, time 7.19s, 
epoch 10, batches 16000|18183, train-acc 0.733, loss 0.641, para-norm 4542.755, grad-norm 3.218, time 7.17s, 
epoch 10, batches 17000|18183, train-acc 0.734, loss 0.641, para-norm 4542.621, grad-norm 1.508, time 7.17s, 
epoch 10, batches 18000|18183, train-acc 0.738, loss 0.634, para-norm 4540.178, grad-norm 1.388, time 7.15s, 
epoch 10, batches 18183|18183, train-acc 0.741, loss 0.626, para-norm 4539.766, grad-norm 6.132, time 1.31s, 
dev-acc 0.767
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
save model!
epoch 11, batches 1000|18183, train-acc 0.735, loss 0.633, para-norm 4542.894, grad-norm 3.374, time 7.46s, 
epoch 11, batches 2000|18183, train-acc 0.736, loss 0.636, para-norm 4545.622, grad-norm 2.666, time 7.73s, 
epoch 11, batches 3000|18183, train-acc 0.736, loss 0.633, para-norm 4546.230, grad-norm 2.822, time 7.78s, 
epoch 11, batches 4000|18183, train-acc 0.737, loss 0.637, para-norm 4544.904, grad-norm 1.026, time 6.91s, 
epoch 11, batches 5000|18183, train-acc 0.735, loss 0.636, para-norm 4547.390, grad-norm 2.623, time 7.04s, 
epoch 11, batches 6000|18183, train-acc 0.737, loss 0.634, para-norm 4546.791, grad-norm 58.113, time 6.94s, 
epoch 11, batches 7000|18183, train-acc 0.732, loss 0.638, para-norm 4547.855, grad-norm 2.161, time 6.95s, 
epoch 11, batches 8000|18183, train-acc 0.738, loss 0.631, para-norm 4549.616, grad-norm 12.162, time 6.94s, 
epoch 11, batches 9000|18183, train-acc 0.739, loss 0.631, para-norm 4551.200, grad-norm 1.807, time 6.94s, 
epoch 11, batches 10000|18183, train-acc 0.744, loss 0.626, para-norm 4552.263, grad-norm 1.907, time 6.95s, 
epoch 11, batches 11000|18183, train-acc 0.736, loss 0.638, para-norm 4554.444, grad-norm 1.125, time 6.99s, 
epoch 11, batches 12000|18183, train-acc 0.736, loss 0.639, para-norm 4559.018, grad-norm 2.847, time 7.00s, 
epoch 11, batches 13000|18183, train-acc 0.740, loss 0.628, para-norm 4561.286, grad-norm 1.312, time 6.96s, 
epoch 11, batches 14000|18183, train-acc 0.738, loss 0.634, para-norm 4560.310, grad-norm 13.949, time 6.98s, 
epoch 11, batches 15000|18183, train-acc 0.738, loss 0.627, para-norm 4561.552, grad-norm 1.686, time 6.95s, 
epoch 11, batches 16000|18183, train-acc 0.743, loss 0.627, para-norm 4560.904, grad-norm 291.494, time 6.99s, 
epoch 11, batches 17000|18183, train-acc 0.743, loss 0.625, para-norm 4561.746, grad-norm 2.957, time 6.95s, 
epoch 11, batches 18000|18183, train-acc 0.742, loss 0.623, para-norm 4564.149, grad-norm 3.804, time 6.94s, 
epoch 11, batches 18183|18183, train-acc 0.752, loss 0.611, para-norm 4564.844, grad-norm 0.803, time 1.27s, 
dev-acc 0.770
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
save model!
epoch 12, batches 1000|18183, train-acc 0.743, loss 0.626, para-norm 4573.376, grad-norm 3.614, time 7.04s, 
epoch 12, batches 2000|18183, train-acc 0.737, loss 0.634, para-norm 4576.130, grad-norm 60.831, time 6.94s, 
epoch 12, batches 3000|18183, train-acc 0.738, loss 0.630, para-norm 4580.390, grad-norm 4.227, time 6.93s, 
epoch 12, batches 4000|18183, train-acc 0.740, loss 0.624, para-norm 4582.544, grad-norm 0.903, time 6.94s, 
epoch 12, batches 5000|18183, train-acc 0.740, loss 0.620, para-norm 4581.097, grad-norm 5.763, time 6.94s, 
epoch 12, batches 6000|18183, train-acc 0.739, loss 0.628, para-norm 4579.892, grad-norm 1.331, time 6.94s, 
epoch 12, batches 7000|18183, train-acc 0.740, loss 0.625, para-norm 4582.630, grad-norm 2.029, time 6.94s, 
epoch 12, batches 8000|18183, train-acc 0.742, loss 0.629, para-norm 4582.253, grad-norm 1.559, time 6.93s, 
epoch 12, batches 9000|18183, train-acc 0.745, loss 0.618, para-norm 4595.662, grad-norm 1.734, time 6.94s, 
epoch 12, batches 10000|18183, train-acc 0.742, loss 0.627, para-norm 4597.813, grad-norm 1.875, time 6.94s, 
epoch 12, batches 11000|18183, train-acc 0.746, loss 0.623, para-norm 4600.437, grad-norm 2.516, time 6.94s, 
epoch 12, batches 12000|18183, train-acc 0.736, loss 0.631, para-norm 4602.421, grad-norm 2.797, time 6.92s, 
epoch 12, batches 13000|18183, train-acc 0.747, loss 0.613, para-norm 4609.615, grad-norm 2.341, time 6.93s, 
epoch 12, batches 14000|18183, train-acc 0.739, loss 0.631, para-norm 4608.925, grad-norm 2.388, time 6.95s, 
epoch 12, batches 15000|18183, train-acc 0.744, loss 0.623, para-norm 4613.784, grad-norm 1.546, time 6.93s, 
epoch 12, batches 16000|18183, train-acc 0.748, loss 0.615, para-norm 4623.676, grad-norm 8.764, time 6.93s, 
epoch 12, batches 17000|18183, train-acc 0.742, loss 0.622, para-norm 4631.283, grad-norm 6.939, time 6.93s, 
epoch 12, batches 18000|18183, train-acc 0.745, loss 0.617, para-norm 4627.594, grad-norm 1.731, time 6.93s, 
epoch 12, batches 18183|18183, train-acc 0.739, loss 0.618, para-norm 4627.276, grad-norm 1.995, time 1.27s, 
dev-acc 0.772
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
save model!
epoch 13, batches 1000|18183, train-acc 0.744, loss 0.620, para-norm 4629.435, grad-norm 1.188, time 7.00s, 
epoch 13, batches 2000|18183, train-acc 0.744, loss 0.621, para-norm 4632.376, grad-norm 1.789, time 6.93s, 
epoch 13, batches 3000|18183, train-acc 0.744, loss 0.620, para-norm 4632.082, grad-norm 0.780, time 6.92s, 
epoch 13, batches 4000|18183, train-acc 0.744, loss 0.613, para-norm 4632.085, grad-norm 1.695, time 6.92s, 
epoch 13, batches 5000|18183, train-acc 0.745, loss 0.616, para-norm 4630.488, grad-norm 1.902, time 6.94s, 
epoch 13, batches 6000|18183, train-acc 0.748, loss 0.615, para-norm 4629.823, grad-norm 1.061, time 6.93s, 
epoch 13, batches 7000|18183, train-acc 0.746, loss 0.618, para-norm 4632.607, grad-norm 1.415, time 6.93s, 
epoch 13, batches 8000|18183, train-acc 0.744, loss 0.619, para-norm 4633.946, grad-norm 2.177, time 6.93s, 
epoch 13, batches 9000|18183, train-acc 0.749, loss 0.613, para-norm 4634.421, grad-norm 3.911, time 6.93s, 
epoch 13, batches 10000|18183, train-acc 0.741, loss 0.625, para-norm 4631.662, grad-norm 53.515, time 6.93s, 
epoch 13, batches 11000|18183, train-acc 0.749, loss 0.612, para-norm 4630.978, grad-norm 3.636, time 6.91s, 
epoch 13, batches 12000|18183, train-acc 0.749, loss 0.617, para-norm 4630.851, grad-norm 1.708, time 6.94s, 
epoch 13, batches 13000|18183, train-acc 0.745, loss 0.616, para-norm 4629.980, grad-norm 2.496, time 6.94s, 
epoch 13, batches 14000|18183, train-acc 0.745, loss 0.621, para-norm 4630.223, grad-norm 3.048, time 6.93s, 
epoch 13, batches 15000|18183, train-acc 0.744, loss 0.620, para-norm 4627.629, grad-norm 50.133, time 6.95s, 
epoch 13, batches 16000|18183, train-acc 0.747, loss 0.612, para-norm 4628.930, grad-norm 2.674, time 6.93s, 
epoch 13, batches 17000|18183, train-acc 0.750, loss 0.608, para-norm 4626.638, grad-norm 3.860, time 6.94s, 
epoch 13, batches 18000|18183, train-acc 0.744, loss 0.615, para-norm 4630.047, grad-norm 5.441, time 6.94s, 
epoch 13, batches 18183|18183, train-acc 0.732, loss 0.639, para-norm 4629.256, grad-norm 3.146, time 1.27s, 
dev-acc 0.774
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
save model!
epoch 14, batches 1000|18183, train-acc 0.746, loss 0.612, para-norm 4629.105, grad-norm 3.897, time 7.05s, 
epoch 14, batches 2000|18183, train-acc 0.744, loss 0.620, para-norm 4626.308, grad-norm 1.551, time 6.92s, 
epoch 14, batches 3000|18183, train-acc 0.749, loss 0.611, para-norm 4633.852, grad-norm 9.305, time 6.94s, 
epoch 14, batches 4000|18183, train-acc 0.746, loss 0.613, para-norm 4637.295, grad-norm 3.681, time 6.94s, 
epoch 14, batches 5000|18183, train-acc 0.752, loss 0.607, para-norm 4648.712, grad-norm 3.300, time 6.95s, 
epoch 14, batches 6000|18183, train-acc 0.746, loss 0.613, para-norm 4647.338, grad-norm 2.382, time 6.93s, 
epoch 14, batches 7000|18183, train-acc 0.749, loss 0.611, para-norm 4650.922, grad-norm 7.639, time 6.93s, 
epoch 14, batches 8000|18183, train-acc 0.749, loss 0.610, para-norm 4648.029, grad-norm 2.342, time 6.92s, 
epoch 14, batches 9000|18183, train-acc 0.744, loss 0.619, para-norm 4648.774, grad-norm 5.773, time 6.94s, 
epoch 14, batches 10000|18183, train-acc 0.747, loss 0.611, para-norm 4647.753, grad-norm 43.460, time 6.94s, 
epoch 14, batches 11000|18183, train-acc 0.754, loss 0.601, para-norm 4648.269, grad-norm 1.037, time 6.92s, 
epoch 14, batches 12000|18183, train-acc 0.752, loss 0.606, para-norm 4648.870, grad-norm 1.704, time 6.99s, 
epoch 14, batches 13000|18183, train-acc 0.749, loss 0.606, para-norm 4648.943, grad-norm 2.349, time 6.93s, 
epoch 14, batches 14000|18183, train-acc 0.753, loss 0.603, para-norm 4645.920, grad-norm 1.573, time 6.94s, 
epoch 14, batches 15000|18183, train-acc 0.747, loss 0.612, para-norm 4653.301, grad-norm 85.333, time 6.91s, 
epoch 14, batches 16000|18183, train-acc 0.746, loss 0.618, para-norm 4651.763, grad-norm 3.672, time 6.93s, 
epoch 14, batches 17000|18183, train-acc 0.750, loss 0.607, para-norm 4649.780, grad-norm 1.769, time 6.93s, 
epoch 14, batches 18000|18183, train-acc 0.748, loss 0.613, para-norm 4646.399, grad-norm 1.616, time 6.93s, 
epoch 14, batches 18183|18183, train-acc 0.753, loss 0.605, para-norm 4646.225, grad-norm 8.066, time 1.27s, 
dev-acc 0.776
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
save model!
epoch 15, batches 1000|18183, train-acc 0.753, loss 0.606, para-norm 4645.980, grad-norm 2.365, time 7.01s, 
epoch 15, batches 2000|18183, train-acc 0.752, loss 0.600, para-norm 4646.066, grad-norm 2.297, time 6.93s, 
epoch 15, batches 3000|18183, train-acc 0.752, loss 0.602, para-norm 4643.363, grad-norm 2.104, time 6.93s, 
epoch 15, batches 4000|18183, train-acc 0.751, loss 0.607, para-norm 4642.702, grad-norm 2.650, time 6.94s, 
epoch 15, batches 5000|18183, train-acc 0.756, loss 0.601, para-norm 4641.819, grad-norm 1.927, time 6.93s, 
epoch 15, batches 6000|18183, train-acc 0.754, loss 0.604, para-norm 4640.150, grad-norm 3.314, time 6.94s, 
epoch 15, batches 7000|18183, train-acc 0.751, loss 0.603, para-norm 4640.190, grad-norm 1.199, time 6.93s, 
epoch 15, batches 8000|18183, train-acc 0.752, loss 0.601, para-norm 4638.975, grad-norm 1.778, time 6.93s, 
epoch 15, batches 9000|18183, train-acc 0.752, loss 0.604, para-norm 4640.069, grad-norm 1.427, time 6.94s, 
epoch 15, batches 10000|18183, train-acc 0.748, loss 0.607, para-norm 4639.710, grad-norm 3.088, time 6.94s, 
epoch 15, batches 11000|18183, train-acc 0.749, loss 0.604, para-norm 4639.519, grad-norm 2.952, time 6.92s, 
epoch 15, batches 12000|18183, train-acc 0.752, loss 0.605, para-norm 4640.397, grad-norm 1.302, time 6.93s, 
epoch 15, batches 13000|18183, train-acc 0.758, loss 0.595, para-norm 4639.284, grad-norm 3.088, time 6.93s, 
epoch 15, batches 14000|18183, train-acc 0.745, loss 0.611, para-norm 4646.059, grad-norm 1.886, time 6.94s, 
epoch 15, batches 15000|18183, train-acc 0.755, loss 0.602, para-norm 4646.427, grad-norm 54.901, time 6.93s, 
epoch 15, batches 16000|18183, train-acc 0.753, loss 0.602, para-norm 4649.407, grad-norm 1.659, time 6.95s, 
epoch 15, batches 17000|18183, train-acc 0.751, loss 0.606, para-norm 4648.740, grad-norm 2.411, time 6.93s, 
epoch 15, batches 18000|18183, train-acc 0.753, loss 0.603, para-norm 4647.938, grad-norm 1.639, time 6.94s, 
epoch 15, batches 18183|18183, train-acc 0.756, loss 0.602, para-norm 4649.592, grad-norm 1.624, time 1.27s, 
dev-acc 0.784
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
save model!
epoch 16, batches 1000|18183, train-acc 0.757, loss 0.595, para-norm 4651.174, grad-norm 4.760, time 7.00s, 
epoch 16, batches 2000|18183, train-acc 0.756, loss 0.597, para-norm 4651.137, grad-norm 2.330, time 6.92s, 
epoch 16, batches 3000|18183, train-acc 0.759, loss 0.592, para-norm 4649.737, grad-norm 2.079, time 6.92s, 
epoch 16, batches 4000|18183, train-acc 0.758, loss 0.591, para-norm 4651.727, grad-norm 12.912, time 6.93s, 
epoch 16, batches 5000|18183, train-acc 0.756, loss 0.597, para-norm 4653.151, grad-norm 2.758, time 6.94s, 
epoch 16, batches 6000|18183, train-acc 0.751, loss 0.608, para-norm 4654.353, grad-norm 14.623, time 6.94s, 
epoch 16, batches 7000|18183, train-acc 0.761, loss 0.585, para-norm 4657.091, grad-norm 3.358, time 6.93s, 
epoch 16, batches 8000|18183, train-acc 0.745, loss 0.610, para-norm 4657.483, grad-norm 1.986, time 6.94s, 
epoch 16, batches 9000|18183, train-acc 0.751, loss 0.599, para-norm 4659.885, grad-norm 1.899, time 6.93s, 
epoch 16, batches 10000|18183, train-acc 0.759, loss 0.591, para-norm 4657.334, grad-norm 34.876, time 6.93s, 
epoch 16, batches 11000|18183, train-acc 0.751, loss 0.601, para-norm 4657.897, grad-norm 5.913, time 6.93s, 
epoch 16, batches 12000|18183, train-acc 0.758, loss 0.595, para-norm 4656.777, grad-norm 3.644, time 6.98s, 
epoch 16, batches 13000|18183, train-acc 0.755, loss 0.597, para-norm 4656.142, grad-norm 2.137, time 6.93s, 
epoch 16, batches 14000|18183, train-acc 0.759, loss 0.593, para-norm 4654.274, grad-norm 1.587, time 6.91s, 
epoch 16, batches 15000|18183, train-acc 0.756, loss 0.596, para-norm 4650.866, grad-norm 1.289, time 6.92s, 
epoch 16, batches 16000|18183, train-acc 0.752, loss 0.602, para-norm 4647.243, grad-norm 2.504, time 6.92s, 
epoch 16, batches 17000|18183, train-acc 0.758, loss 0.594, para-norm 4645.535, grad-norm 1.900, time 6.92s, 
epoch 16, batches 18000|18183, train-acc 0.754, loss 0.600, para-norm 4645.525, grad-norm 4.076, time 6.93s, 
epoch 16, batches 18183|18183, train-acc 0.756, loss 0.596, para-norm 4646.038, grad-norm 4.222, time 1.26s, 
dev-acc 0.781
epoch 17, batches 1000|18183, train-acc 0.762, loss 0.586, para-norm 4644.288, grad-norm 2.842, time 7.02s, 
epoch 17, batches 2000|18183, train-acc 0.755, loss 0.596, para-norm 4640.876, grad-norm 2758.078, time 6.92s, 
epoch 17, batches 3000|18183, train-acc 0.761, loss 0.588, para-norm 4643.483, grad-norm 1.895, time 6.90s, 
epoch 17, batches 4000|18183, train-acc 0.759, loss 0.588, para-norm 4644.587, grad-norm 2.045, time 6.93s, 
epoch 17, batches 5000|18183, train-acc 0.758, loss 0.594, para-norm 4647.535, grad-norm 1.697, time 6.92s, 
epoch 17, batches 6000|18183, train-acc 0.756, loss 0.595, para-norm 4655.832, grad-norm 1.845, time 6.92s, 
epoch 17, batches 7000|18183, train-acc 0.753, loss 0.601, para-norm 4659.820, grad-norm 10.315, time 6.92s, 
epoch 17, batches 8000|18183, train-acc 0.755, loss 0.599, para-norm 4659.386, grad-norm 2.356, time 6.96s, 
epoch 17, batches 9000|18183, train-acc 0.757, loss 0.596, para-norm 4665.461, grad-norm 1.713, time 6.91s, 
epoch 17, batches 10000|18183, train-acc 0.755, loss 0.594, para-norm 4662.727, grad-norm 34.410, time 6.92s, 
epoch 17, batches 11000|18183, train-acc 0.760, loss 0.591, para-norm 4665.481, grad-norm 3.032, time 6.98s, 
epoch 17, batches 12000|18183, train-acc 0.758, loss 0.588, para-norm 4664.027, grad-norm 1.711, time 6.93s, 
epoch 17, batches 13000|18183, train-acc 0.761, loss 0.589, para-norm 4662.284, grad-norm 3.649, time 6.92s, 
epoch 17, batches 14000|18183, train-acc 0.760, loss 0.591, para-norm 4662.370, grad-norm 2.990, time 6.93s, 
epoch 17, batches 15000|18183, train-acc 0.759, loss 0.591, para-norm 4661.310, grad-norm 3.965, time 6.92s, 
epoch 17, batches 16000|18183, train-acc 0.760, loss 0.586, para-norm 4659.894, grad-norm 2.283, time 6.93s, 
epoch 17, batches 17000|18183, train-acc 0.757, loss 0.592, para-norm 4657.093, grad-norm 2.939, time 6.95s, 
epoch 17, batches 18000|18183, train-acc 0.757, loss 0.591, para-norm 4658.464, grad-norm 12.793, time 6.94s, 
epoch 17, batches 18183|18183, train-acc 0.752, loss 0.596, para-norm 4657.674, grad-norm 2.447, time 1.27s, 
dev-acc 0.789
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
save model!
epoch 18, batches 1000|18183, train-acc 0.764, loss 0.585, para-norm 4656.758, grad-norm 2.616, time 7.02s, 
epoch 18, batches 2000|18183, train-acc 0.758, loss 0.590, para-norm 4654.430, grad-norm 2.414, time 6.92s, 
epoch 18, batches 3000|18183, train-acc 0.758, loss 0.587, para-norm 4655.232, grad-norm 3.233, time 6.93s, 
epoch 18, batches 4000|18183, train-acc 0.760, loss 0.585, para-norm 4656.429, grad-norm 2.000, time 6.92s, 
epoch 18, batches 5000|18183, train-acc 0.765, loss 0.573, para-norm 4656.884, grad-norm 5.078, time 6.92s, 
epoch 18, batches 6000|18183, train-acc 0.760, loss 0.586, para-norm 4660.192, grad-norm 7.959, time 6.93s, 
epoch 18, batches 7000|18183, train-acc 0.760, loss 0.588, para-norm 4658.113, grad-norm 3.210, time 6.95s, 
epoch 18, batches 8000|18183, train-acc 0.756, loss 0.591, para-norm 4664.145, grad-norm 3.577, time 6.94s, 
epoch 18, batches 9000|18183, train-acc 0.757, loss 0.595, para-norm 4663.079, grad-norm 2.139, time 6.93s, 
epoch 18, batches 10000|18183, train-acc 0.758, loss 0.589, para-norm 4664.492, grad-norm 1.315, time 6.93s, 
epoch 18, batches 11000|18183, train-acc 0.760, loss 0.583, para-norm 4663.720, grad-norm 3.560, time 6.94s, 
epoch 18, batches 12000|18183, train-acc 0.756, loss 0.590, para-norm 4661.530, grad-norm 2.333, time 6.92s, 
epoch 18, batches 13000|18183, train-acc 0.762, loss 0.584, para-norm 4659.310, grad-norm 2.184, time 6.92s, 
epoch 18, batches 14000|18183, train-acc 0.758, loss 0.589, para-norm 4660.470, grad-norm 1.688, time 6.93s, 
epoch 18, batches 15000|18183, train-acc 0.757, loss 0.589, para-norm 4658.395, grad-norm 7.028, time 6.93s, 
epoch 18, batches 16000|18183, train-acc 0.760, loss 0.586, para-norm 4655.740, grad-norm 2.885, time 6.94s, 
epoch 18, batches 17000|18183, train-acc 0.758, loss 0.593, para-norm 4656.573, grad-norm 4.278, time 6.92s, 
epoch 18, batches 18000|18183, train-acc 0.755, loss 0.597, para-norm 4656.163, grad-norm 2.587, time 7.01s, 
epoch 18, batches 18183|18183, train-acc 0.759, loss 0.586, para-norm 4654.909, grad-norm 2.212, time 1.26s, 
dev-acc 0.788
epoch 19, batches 1000|18183, train-acc 0.760, loss 0.587, para-norm 4655.028, grad-norm 1.145, time 7.04s, 
epoch 19, batches 2000|18183, train-acc 0.761, loss 0.587, para-norm 4652.735, grad-norm 2.567, time 6.93s, 
epoch 19, batches 3000|18183, train-acc 0.761, loss 0.584, para-norm 4656.374, grad-norm 1.456, time 6.92s, 
epoch 19, batches 4000|18183, train-acc 0.760, loss 0.582, para-norm 4655.727, grad-norm 3.080, time 6.92s, 
epoch 19, batches 5000|18183, train-acc 0.763, loss 0.584, para-norm 4654.642, grad-norm 2.615, time 6.91s, 
epoch 19, batches 6000|18183, train-acc 0.759, loss 0.582, para-norm 4654.696, grad-norm 4.730, time 6.94s, 
epoch 19, batches 7000|18183, train-acc 0.757, loss 0.596, para-norm 4655.685, grad-norm 4.080, time 6.93s, 
epoch 19, batches 8000|18183, train-acc 0.757, loss 0.592, para-norm 4653.373, grad-norm 0.932, time 6.91s, 
epoch 19, batches 9000|18183, train-acc 0.765, loss 0.579, para-norm 4651.993, grad-norm 2.985, time 6.92s, 
epoch 19, batches 10000|18183, train-acc 0.763, loss 0.587, para-norm 4652.192, grad-norm 21.365, time 6.91s, 
epoch 19, batches 11000|18183, train-acc 0.762, loss 0.578, para-norm 4651.891, grad-norm 3.155, time 6.93s, 
epoch 19, batches 12000|18183, train-acc 0.764, loss 0.588, para-norm 4662.720, grad-norm 2.144, time 6.94s, 
epoch 19, batches 13000|18183, train-acc 0.758, loss 0.586, para-norm 4663.281, grad-norm 4.044, time 6.92s, 
epoch 19, batches 14000|18183, train-acc 0.759, loss 0.587, para-norm 4665.342, grad-norm 1.731, time 6.92s, 
epoch 19, batches 15000|18183, train-acc 0.765, loss 0.577, para-norm 4668.827, grad-norm 3.044, time 6.94s, 
epoch 19, batches 16000|18183, train-acc 0.761, loss 0.583, para-norm 4669.490, grad-norm 1.563, time 6.92s, 
epoch 19, batches 17000|18183, train-acc 0.765, loss 0.575, para-norm 4669.694, grad-norm 2.359, time 6.93s, 
epoch 19, batches 18000|18183, train-acc 0.765, loss 0.579, para-norm 4674.569, grad-norm 3.787, time 6.93s, 
epoch 19, batches 18183|18183, train-acc 0.760, loss 0.592, para-norm 4674.856, grad-norm 1.832, time 1.27s, 
dev-acc 0.790
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
save model!
epoch 20, batches 1000|18183, train-acc 0.765, loss 0.575, para-norm 4674.485, grad-norm 59.693, time 7.04s, 
epoch 20, batches 2000|18183, train-acc 0.759, loss 0.582, para-norm 4675.197, grad-norm 1.920, time 6.93s, 
epoch 20, batches 3000|18183, train-acc 0.764, loss 0.577, para-norm 4676.123, grad-norm 1.703, time 6.92s, 
epoch 20, batches 4000|18183, train-acc 0.768, loss 0.571, para-norm 4673.348, grad-norm 2.293, time 6.94s, 
epoch 20, batches 5000|18183, train-acc 0.761, loss 0.588, para-norm 4675.472, grad-norm 2.378, time 6.93s, 
epoch 20, batches 6000|18183, train-acc 0.760, loss 0.583, para-norm 4676.917, grad-norm 2.844, time 6.93s, 
epoch 20, batches 7000|18183, train-acc 0.760, loss 0.583, para-norm 4675.313, grad-norm 3.119, time 6.93s, 
epoch 20, batches 8000|18183, train-acc 0.760, loss 0.581, para-norm 4673.459, grad-norm 2.121, time 6.91s, 
epoch 20, batches 9000|18183, train-acc 0.760, loss 0.584, para-norm 4672.307, grad-norm 2.495, time 6.93s, 
epoch 20, batches 10000|18183, train-acc 0.763, loss 0.577, para-norm 4670.299, grad-norm 2.662, time 6.91s, 
epoch 20, batches 11000|18183, train-acc 0.764, loss 0.582, para-norm 4668.484, grad-norm 6.987, time 6.93s, 
epoch 20, batches 12000|18183, train-acc 0.766, loss 0.577, para-norm 4671.000, grad-norm 4.542, time 6.92s, 
epoch 20, batches 13000|18183, train-acc 0.762, loss 0.581, para-norm 4668.704, grad-norm 1.645, time 6.93s, 
epoch 20, batches 14000|18183, train-acc 0.762, loss 0.583, para-norm 4675.345, grad-norm 4.415, time 6.92s, 
epoch 20, batches 15000|18183, train-acc 0.767, loss 0.577, para-norm 4682.121, grad-norm 3.538, time 6.93s, 
epoch 20, batches 16000|18183, train-acc 0.761, loss 0.583, para-norm 4678.671, grad-norm 3.655, time 6.93s, 
epoch 20, batches 17000|18183, train-acc 0.766, loss 0.575, para-norm 4677.337, grad-norm 29.369, time 6.94s, 
epoch 20, batches 18000|18183, train-acc 0.763, loss 0.584, para-norm 4675.463, grad-norm 4.051, time 6.94s, 
epoch 20, batches 18183|18183, train-acc 0.761, loss 0.595, para-norm 4675.084, grad-norm 10.087, time 1.27s, 
dev-acc 0.793
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
save model!
epoch 21, batches 1000|18183, train-acc 0.772, loss 0.567, para-norm 4676.206, grad-norm 2.035, time 6.99s, 
epoch 21, batches 2000|18183, train-acc 0.765, loss 0.576, para-norm 4676.142, grad-norm 2.571, time 6.92s, 
epoch 21, batches 3000|18183, train-acc 0.761, loss 0.581, para-norm 4676.658, grad-norm 7.314, time 6.93s, 
epoch 21, batches 4000|18183, train-acc 0.770, loss 0.566, para-norm 4682.696, grad-norm 2.478, time 6.93s, 
epoch 21, batches 5000|18183, train-acc 0.768, loss 0.575, para-norm 4683.544, grad-norm 3.007, time 6.93s, 
epoch 21, batches 6000|18183, train-acc 0.763, loss 0.578, para-norm 4683.047, grad-norm 5.194, time 6.94s, 
epoch 21, batches 7000|18183, train-acc 0.765, loss 0.578, para-norm 4682.004, grad-norm 24.157, time 6.94s, 
epoch 21, batches 8000|18183, train-acc 0.761, loss 0.579, para-norm 4682.787, grad-norm 3.045, time 6.92s, 
epoch 21, batches 9000|18183, train-acc 0.765, loss 0.575, para-norm 4690.397, grad-norm 3.208, time 6.92s, 
epoch 21, batches 10000|18183, train-acc 0.763, loss 0.576, para-norm 4691.221, grad-norm 2.393, time 6.93s, 
epoch 21, batches 11000|18183, train-acc 0.763, loss 0.581, para-norm 4697.639, grad-norm 2.380, time 6.93s, 
epoch 21, batches 12000|18183, train-acc 0.764, loss 0.576, para-norm 4701.225, grad-norm 2.332, time 6.93s, 
epoch 21, batches 13000|18183, train-acc 0.763, loss 0.577, para-norm 4702.285, grad-norm 7.484, time 6.93s, 
epoch 21, batches 14000|18183, train-acc 0.769, loss 0.569, para-norm 4704.200, grad-norm 3.811, time 6.93s, 
epoch 21, batches 15000|18183, train-acc 0.764, loss 0.573, para-norm 4708.072, grad-norm 1.794, time 6.94s, 
epoch 21, batches 16000|18183, train-acc 0.762, loss 0.583, para-norm 4708.562, grad-norm 2.178, time 6.93s, 
epoch 21, batches 17000|18183, train-acc 0.770, loss 0.571, para-norm 4706.822, grad-norm 2.755, time 6.93s, 
epoch 21, batches 18000|18183, train-acc 0.765, loss 0.576, para-norm 4707.452, grad-norm 3.067, time 6.93s, 
epoch 21, batches 18183|18183, train-acc 0.773, loss 0.557, para-norm 4707.297, grad-norm 2.467, time 1.27s, 
dev-acc 0.793
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
save model!
epoch 22, batches 1000|18183, train-acc 0.768, loss 0.567, para-norm 4706.222, grad-norm 3.484, time 7.00s, 
epoch 22, batches 2000|18183, train-acc 0.762, loss 0.577, para-norm 4704.179, grad-norm 3.177, time 6.93s, 
epoch 22, batches 3000|18183, train-acc 0.770, loss 0.565, para-norm 4705.270, grad-norm 2.429, time 6.93s, 
epoch 22, batches 4000|18183, train-acc 0.761, loss 0.583, para-norm 4704.385, grad-norm 2.166, time 6.94s, 
epoch 22, batches 5000|18183, train-acc 0.762, loss 0.583, para-norm 4703.815, grad-norm 2.741, time 6.92s, 
epoch 22, batches 6000|18183, train-acc 0.767, loss 0.571, para-norm 4701.994, grad-norm 2.116, time 6.94s, 
epoch 22, batches 7000|18183, train-acc 0.770, loss 0.569, para-norm 4701.499, grad-norm 2.495, time 6.93s, 
epoch 22, batches 8000|18183, train-acc 0.770, loss 0.571, para-norm 4698.887, grad-norm 2.760, time 6.93s, 
epoch 22, batches 9000|18183, train-acc 0.764, loss 0.581, para-norm 4697.890, grad-norm 4.235, time 6.93s, 
epoch 22, batches 10000|18183, train-acc 0.768, loss 0.569, para-norm 4697.601, grad-norm 3.965, time 6.93s, 
epoch 22, batches 11000|18183, train-acc 0.763, loss 0.579, para-norm 4695.306, grad-norm 2.062, time 6.93s, 
epoch 22, batches 12000|18183, train-acc 0.766, loss 0.576, para-norm 4693.363, grad-norm 1.837, time 6.92s, 
epoch 22, batches 13000|18183, train-acc 0.770, loss 0.565, para-norm 4694.611, grad-norm 2.826, time 6.93s, 
epoch 22, batches 14000|18183, train-acc 0.769, loss 0.567, para-norm 4695.706, grad-norm 1.687, time 6.93s, 
epoch 22, batches 15000|18183, train-acc 0.767, loss 0.572, para-norm 4698.276, grad-norm 1.315, time 6.94s, 
epoch 22, batches 16000|18183, train-acc 0.769, loss 0.569, para-norm 4697.372, grad-norm 146.448, time 6.94s, 
epoch 22, batches 17000|18183, train-acc 0.764, loss 0.577, para-norm 4695.206, grad-norm 3.251, time 6.93s, 
epoch 22, batches 18000|18183, train-acc 0.771, loss 0.565, para-norm 4698.098, grad-norm 1.378, time 6.91s, 
epoch 22, batches 18183|18183, train-acc 0.762, loss 0.571, para-norm 4696.986, grad-norm 2.084, time 1.27s, 
dev-acc 0.794
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
save model!
epoch 23, batches 1000|18183, train-acc 0.767, loss 0.571, para-norm 4695.919, grad-norm 1.956, time 7.07s, 
epoch 23, batches 2000|18183, train-acc 0.768, loss 0.572, para-norm 4699.845, grad-norm 24.774, time 6.94s, 
epoch 23, batches 3000|18183, train-acc 0.768, loss 0.564, para-norm 4707.308, grad-norm 1.983, time 6.94s, 
epoch 23, batches 4000|18183, train-acc 0.771, loss 0.566, para-norm 4712.341, grad-norm 2.658, time 6.94s, 
epoch 23, batches 5000|18183, train-acc 0.766, loss 0.572, para-norm 4711.187, grad-norm 2.830, time 6.94s, 
epoch 23, batches 6000|18183, train-acc 0.770, loss 0.565, para-norm 4708.357, grad-norm 3.182, time 6.95s, 
epoch 23, batches 7000|18183, train-acc 0.767, loss 0.572, para-norm 4708.398, grad-norm 2.843, time 6.93s, 
epoch 23, batches 8000|18183, train-acc 0.769, loss 0.564, para-norm 4708.107, grad-norm 15.519, time 6.93s, 
epoch 23, batches 9000|18183, train-acc 0.768, loss 0.570, para-norm 4713.184, grad-norm 3.593, time 6.93s, 
epoch 23, batches 10000|18183, train-acc 0.773, loss 0.562, para-norm 4713.758, grad-norm 3.010, time 6.93s, 
epoch 23, batches 11000|18183, train-acc 0.768, loss 0.567, para-norm 4712.600, grad-norm 1.742, time 6.94s, 
epoch 23, batches 12000|18183, train-acc 0.769, loss 0.567, para-norm 4711.971, grad-norm 5.302, time 6.92s, 
epoch 23, batches 13000|18183, train-acc 0.765, loss 0.575, para-norm 4709.075, grad-norm 1.634, time 6.93s, 
epoch 23, batches 14000|18183, train-acc 0.769, loss 0.572, para-norm 4706.863, grad-norm 0.676, time 6.93s, 
epoch 23, batches 15000|18183, train-acc 0.766, loss 0.569, para-norm 4708.148, grad-norm 5336.816, time 6.94s, 
epoch 23, batches 16000|18183, train-acc 0.763, loss 0.577, para-norm 4708.000, grad-norm 1.277, time 6.94s, 
epoch 23, batches 17000|18183, train-acc 0.771, loss 0.560, para-norm 4703.945, grad-norm 2.609, time 6.94s, 
epoch 23, batches 18000|18183, train-acc 0.766, loss 0.572, para-norm 4704.489, grad-norm 5.015, time 6.94s, 
epoch 23, batches 18183|18183, train-acc 0.775, loss 0.565, para-norm 4705.080, grad-norm 3.150, time 1.27s, 
dev-acc 0.793
epoch 24, batches 1000|18183, train-acc 0.765, loss 0.575, para-norm 4708.627, grad-norm 9.270, time 7.01s, 
epoch 24, batches 2000|18183, train-acc 0.768, loss 0.572, para-norm 4717.505, grad-norm 1.583, time 6.93s, 
epoch 24, batches 3000|18183, train-acc 0.769, loss 0.568, para-norm 4717.565, grad-norm 3.882, time 6.94s, 
epoch 24, batches 4000|18183, train-acc 0.776, loss 0.559, para-norm 4725.016, grad-norm 11.394, time 6.95s, 
epoch 24, batches 5000|18183, train-acc 0.769, loss 0.566, para-norm 4726.132, grad-norm 2.502, time 6.95s, 
epoch 24, batches 6000|18183, train-acc 0.773, loss 0.559, para-norm 4726.720, grad-norm 3.712, time 6.94s, 
epoch 24, batches 7000|18183, train-acc 0.770, loss 0.566, para-norm 4727.943, grad-norm 1.477, time 6.96s, 
epoch 24, batches 8000|18183, train-acc 0.770, loss 0.564, para-norm 4728.189, grad-norm 2.797, time 6.94s, 
epoch 24, batches 9000|18183, train-acc 0.773, loss 0.562, para-norm 4725.781, grad-norm 1.483, time 6.93s, 
epoch 24, batches 10000|18183, train-acc 0.770, loss 0.566, para-norm 4724.717, grad-norm 250.245, time 6.93s, 
epoch 24, batches 11000|18183, train-acc 0.770, loss 0.566, para-norm 4722.231, grad-norm 2.297, time 6.95s, 
epoch 24, batches 12000|18183, train-acc 0.773, loss 0.564, para-norm 4726.871, grad-norm 1.411, time 6.95s, 
epoch 24, batches 13000|18183, train-acc 0.775, loss 0.557, para-norm 4727.971, grad-norm 2.093, time 6.94s, 
epoch 24, batches 14000|18183, train-acc 0.769, loss 0.565, para-norm 4728.094, grad-norm 2.661, time 6.95s, 
epoch 24, batches 15000|18183, train-acc 0.769, loss 0.569, para-norm 4728.977, grad-norm 1.220, time 6.94s, 
epoch 24, batches 16000|18183, train-acc 0.773, loss 0.565, para-norm 4728.821, grad-norm 2.906, time 6.94s, 
epoch 24, batches 17000|18183, train-acc 0.769, loss 0.572, para-norm 4730.086, grad-norm 3.853, time 6.95s, 
epoch 24, batches 18000|18183, train-acc 0.763, loss 0.575, para-norm 4729.799, grad-norm 3.424, time 6.94s, 
epoch 24, batches 18183|18183, train-acc 0.775, loss 0.559, para-norm 4730.642, grad-norm 3.639, time 1.27s, 
dev-acc 0.795
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
save model!
epoch 25, batches 1000|18183, train-acc 0.774, loss 0.557, para-norm 4727.306, grad-norm 2.209, time 7.01s, 
epoch 25, batches 2000|18183, train-acc 0.768, loss 0.570, para-norm 4726.109, grad-norm 3.932, time 6.94s, 
epoch 25, batches 3000|18183, train-acc 0.773, loss 0.561, para-norm 4725.991, grad-norm 7.238, time 6.94s, 
epoch 25, batches 4000|18183, train-acc 0.771, loss 0.561, para-norm 4724.849, grad-norm 2.911, time 6.94s, 
epoch 25, batches 5000|18183, train-acc 0.771, loss 0.564, para-norm 4725.079, grad-norm 45.070, time 6.94s, 
epoch 25, batches 6000|18183, train-acc 0.770, loss 0.564, para-norm 4725.680, grad-norm 2.645, time 6.95s, 
epoch 25, batches 7000|18183, train-acc 0.775, loss 0.559, para-norm 4725.749, grad-norm 2.511, time 6.94s, 
epoch 25, batches 8000|18183, train-acc 0.765, loss 0.575, para-norm 4725.737, grad-norm 5.261, time 6.95s, 
epoch 25, batches 9000|18183, train-acc 0.771, loss 0.563, para-norm 4723.503, grad-norm 2.815, time 6.95s, 
epoch 25, batches 10000|18183, train-acc 0.772, loss 0.565, para-norm 4724.270, grad-norm 2.137, time 6.95s, 
epoch 25, batches 11000|18183, train-acc 0.775, loss 0.556, para-norm 4722.136, grad-norm 4.030, time 6.93s, 
epoch 25, batches 12000|18183, train-acc 0.770, loss 0.569, para-norm 4721.901, grad-norm 2.783, time 6.94s, 
epoch 25, batches 13000|18183, train-acc 0.771, loss 0.558, para-norm 4728.648, grad-norm 2.671, time 6.95s, 
epoch 25, batches 14000|18183, train-acc 0.773, loss 0.557, para-norm 4728.016, grad-norm 3.277, time 6.94s, 
epoch 25, batches 15000|18183, train-acc 0.775, loss 0.560, para-norm 4726.586, grad-norm 2.129, time 6.95s, 
epoch 25, batches 16000|18183, train-acc 0.772, loss 0.563, para-norm 4727.648, grad-norm 2.264, time 6.95s, 
epoch 25, batches 17000|18183, train-acc 0.773, loss 0.558, para-norm 4725.583, grad-norm 6.931, time 6.94s, 
epoch 25, batches 18000|18183, train-acc 0.772, loss 0.562, para-norm 4725.693, grad-norm 4.197, time 6.95s, 
epoch 25, batches 18183|18183, train-acc 0.768, loss 0.569, para-norm 4724.602, grad-norm 2.759, time 1.27s, 
dev-acc 0.794
epoch 26, batches 1000|18183, train-acc 0.776, loss 0.555, para-norm 4724.603, grad-norm 2.936, time 7.03s, 
epoch 26, batches 2000|18183, train-acc 0.774, loss 0.558, para-norm 4725.852, grad-norm 2.326, time 6.95s, 
epoch 26, batches 3000|18183, train-acc 0.772, loss 0.559, para-norm 4727.031, grad-norm 1.511, time 6.93s, 
epoch 26, batches 4000|18183, train-acc 0.773, loss 0.564, para-norm 4736.416, grad-norm 2.182, time 6.95s, 
epoch 26, batches 5000|18183, train-acc 0.771, loss 0.562, para-norm 4737.120, grad-norm 22.207, time 6.94s, 
epoch 26, batches 6000|18183, train-acc 0.777, loss 0.561, para-norm 4736.423, grad-norm 1.095, time 6.95s, 
epoch 26, batches 7000|18183, train-acc 0.771, loss 0.565, para-norm 4736.812, grad-norm 5.762, time 6.95s, 
epoch 26, batches 8000|18183, train-acc 0.773, loss 0.560, para-norm 4736.628, grad-norm 1.900, time 6.95s, 
epoch 26, batches 9000|18183, train-acc 0.773, loss 0.555, para-norm 4736.560, grad-norm 1.163, time 6.95s, 
epoch 26, batches 10000|18183, train-acc 0.765, loss 0.570, para-norm 4735.278, grad-norm 4.966, time 6.96s, 
epoch 26, batches 11000|18183, train-acc 0.774, loss 0.558, para-norm 4733.182, grad-norm 5.092, time 6.96s, 
epoch 26, batches 12000|18183, train-acc 0.775, loss 0.559, para-norm 4730.971, grad-norm 2.689, time 6.95s, 
epoch 26, batches 13000|18183, train-acc 0.770, loss 0.564, para-norm 4730.790, grad-norm 2.407, time 6.96s, 
epoch 26, batches 14000|18183, train-acc 0.772, loss 0.557, para-norm 4728.929, grad-norm 1.561, time 6.95s, 
epoch 26, batches 15000|18183, train-acc 0.777, loss 0.554, para-norm 4735.087, grad-norm 3.503, time 7.01s, 
epoch 26, batches 16000|18183, train-acc 0.773, loss 0.558, para-norm 4734.151, grad-norm 15.476, time 6.93s, 
epoch 26, batches 17000|18183, train-acc 0.771, loss 0.561, para-norm 4733.350, grad-norm 16.811, time 6.94s, 
epoch 26, batches 18000|18183, train-acc 0.775, loss 0.555, para-norm 4732.921, grad-norm 23.591, time 6.94s, 
epoch 26, batches 18183|18183, train-acc 0.788, loss 0.542, para-norm 4731.585, grad-norm 2.214, time 1.27s, 
dev-acc 0.802
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
save model!
epoch 27, batches 1000|18183, train-acc 0.777, loss 0.551, para-norm 4734.868, grad-norm 28.227, time 7.06s, 
epoch 27, batches 2000|18183, train-acc 0.773, loss 0.558, para-norm 4732.754, grad-norm 3.345, time 7.25s, 
epoch 27, batches 3000|18183, train-acc 0.775, loss 0.556, para-norm 4730.815, grad-norm 2.307, time 7.76s, 
epoch 27, batches 4000|18183, train-acc 0.775, loss 0.560, para-norm 4730.348, grad-norm 0.850, time 7.77s, 
epoch 27, batches 5000|18183, train-acc 0.773, loss 0.562, para-norm 4728.082, grad-norm 0.969, time 7.78s, 
epoch 27, batches 6000|18183, train-acc 0.775, loss 0.556, para-norm 4729.967, grad-norm 3.037, time 7.71s, 
epoch 27, batches 7000|18183, train-acc 0.775, loss 0.555, para-norm 4729.405, grad-norm 3.154, time 6.93s, 
epoch 27, batches 8000|18183, train-acc 0.779, loss 0.551, para-norm 4726.012, grad-norm 11.458, time 6.92s, 
epoch 27, batches 9000|18183, train-acc 0.777, loss 0.553, para-norm 4723.753, grad-norm 3.724, time 6.92s, 
epoch 27, batches 10000|18183, train-acc 0.767, loss 0.565, para-norm 4722.166, grad-norm 2.594, time 6.94s, 
epoch 27, batches 11000|18183, train-acc 0.773, loss 0.556, para-norm 4723.659, grad-norm 4.595, time 6.94s, 
epoch 27, batches 12000|18183, train-acc 0.771, loss 0.563, para-norm 4723.002, grad-norm 3.804, time 6.94s, 
epoch 27, batches 13000|18183, train-acc 0.772, loss 0.563, para-norm 4722.718, grad-norm 4.330, time 7.01s, 
epoch 27, batches 14000|18183, train-acc 0.773, loss 0.562, para-norm 4720.544, grad-norm 3.119, time 6.95s, 
epoch 27, batches 15000|18183, train-acc 0.780, loss 0.547, para-norm 4724.695, grad-norm 2.469, time 6.95s, 
epoch 27, batches 16000|18183, train-acc 0.772, loss 0.557, para-norm 4722.465, grad-norm 33.255, time 6.96s, 
epoch 27, batches 17000|18183, train-acc 0.777, loss 0.550, para-norm 4721.174, grad-norm 1.314, time 6.94s, 
epoch 27, batches 18000|18183, train-acc 0.774, loss 0.554, para-norm 4721.769, grad-norm 2.833, time 6.94s, 
epoch 27, batches 18183|18183, train-acc 0.777, loss 0.555, para-norm 4722.696, grad-norm 2.444, time 1.27s, 
dev-acc 0.802
epoch 28, batches 1000|18183, train-acc 0.781, loss 0.545, para-norm 4725.139, grad-norm 0.002, time 7.10s, 
epoch 28, batches 2000|18183, train-acc 0.776, loss 0.556, para-norm 4730.109, grad-norm 2.589, time 6.96s, 
epoch 28, batches 3000|18183, train-acc 0.775, loss 0.554, para-norm 4733.325, grad-norm 52.331, time 6.96s, 
epoch 28, batches 4000|18183, train-acc 0.774, loss 0.558, para-norm 4733.471, grad-norm 3.360, time 6.93s, 
epoch 28, batches 5000|18183, train-acc 0.775, loss 0.557, para-norm 4731.396, grad-norm 2.320, time 6.95s, 
epoch 28, batches 6000|18183, train-acc 0.776, loss 0.553, para-norm 4731.309, grad-norm 4.906, time 6.96s, 
epoch 28, batches 7000|18183, train-acc 0.774, loss 0.553, para-norm 4729.943, grad-norm 1.475, time 6.94s, 
epoch 28, batches 8000|18183, train-acc 0.779, loss 0.545, para-norm 4726.830, grad-norm 860.380, time 6.95s, 
epoch 28, batches 9000|18183, train-acc 0.772, loss 0.558, para-norm 4727.630, grad-norm 2.211, time 6.95s, 
epoch 28, batches 10000|18183, train-acc 0.775, loss 0.559, para-norm 4725.818, grad-norm 2.692, time 6.94s, 
epoch 28, batches 11000|18183, train-acc 0.772, loss 0.559, para-norm 4725.966, grad-norm 1.651, time 7.01s, 
epoch 28, batches 12000|18183, train-acc 0.775, loss 0.555, para-norm 4725.389, grad-norm 3.471, time 6.96s, 
epoch 28, batches 13000|18183, train-acc 0.775, loss 0.551, para-norm 4732.938, grad-norm 3.707, time 6.94s, 
epoch 28, batches 14000|18183, train-acc 0.777, loss 0.553, para-norm 4730.846, grad-norm 3.346, time 6.94s, 
epoch 28, batches 15000|18183, train-acc 0.777, loss 0.552, para-norm 4730.387, grad-norm 1.450, time 6.94s, 
epoch 28, batches 16000|18183, train-acc 0.777, loss 0.551, para-norm 4738.199, grad-norm 3.040, time 6.95s, 
epoch 28, batches 17000|18183, train-acc 0.773, loss 0.558, para-norm 4739.939, grad-norm 5.011, time 6.96s, 
epoch 28, batches 18000|18183, train-acc 0.776, loss 0.556, para-norm 4739.564, grad-norm 4.811, time 6.96s, 
epoch 28, batches 18183|18183, train-acc 0.780, loss 0.545, para-norm 4739.843, grad-norm 3.388, time 1.28s, 
dev-acc 0.806
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
save model!
epoch 29, batches 1000|18183, train-acc 0.782, loss 0.543, para-norm 4740.927, grad-norm 3480.351, time 7.01s, 
epoch 29, batches 2000|18183, train-acc 0.776, loss 0.551, para-norm 4742.180, grad-norm 6.718, time 6.97s, 
epoch 29, batches 3000|18183, train-acc 0.776, loss 0.554, para-norm 4742.306, grad-norm 1.595, time 6.95s, 
epoch 29, batches 4000|18183, train-acc 0.778, loss 0.548, para-norm 4742.858, grad-norm 3.130, time 6.94s, 
epoch 29, batches 5000|18183, train-acc 0.778, loss 0.549, para-norm 4742.434, grad-norm 4.194, time 6.93s, 
epoch 29, batches 6000|18183, train-acc 0.774, loss 0.557, para-norm 4740.825, grad-norm 3.913, time 6.95s, 
epoch 29, batches 7000|18183, train-acc 0.774, loss 0.554, para-norm 4741.323, grad-norm 1.831, time 6.95s, 
epoch 29, batches 8000|18183, train-acc 0.779, loss 0.546, para-norm 4738.889, grad-norm 1.510, time 6.95s, 
epoch 29, batches 9000|18183, train-acc 0.776, loss 0.550, para-norm 4737.225, grad-norm 2.066, time 6.95s, 
epoch 29, batches 10000|18183, train-acc 0.774, loss 0.556, para-norm 4735.923, grad-norm 18.217, time 6.94s, 
epoch 29, batches 11000|18183, train-acc 0.778, loss 0.554, para-norm 4745.043, grad-norm 3.277, time 6.95s, 
epoch 29, batches 12000|18183, train-acc 0.772, loss 0.559, para-norm 4742.576, grad-norm 2.259, time 6.95s, 
epoch 29, batches 13000|18183, train-acc 0.779, loss 0.548, para-norm 4746.420, grad-norm 2.814, time 6.93s, 
epoch 29, batches 14000|18183, train-acc 0.778, loss 0.553, para-norm 4744.861, grad-norm 1.523, time 6.95s, 
epoch 29, batches 15000|18183, train-acc 0.775, loss 0.557, para-norm 4745.181, grad-norm 8.265, time 6.95s, 
epoch 29, batches 16000|18183, train-acc 0.778, loss 0.547, para-norm 4745.276, grad-norm 4.513, time 6.95s, 
epoch 29, batches 17000|18183, train-acc 0.774, loss 0.554, para-norm 4745.997, grad-norm 1.849, time 6.96s, 
epoch 29, batches 18000|18183, train-acc 0.772, loss 0.559, para-norm 4745.053, grad-norm 2.548, time 6.95s, 
epoch 29, batches 18183|18183, train-acc 0.781, loss 0.539, para-norm 4744.927, grad-norm 16.394, time 1.27s, 
dev-acc 0.801
epoch 30, batches 1000|18183, train-acc 0.779, loss 0.547, para-norm 4744.702, grad-norm 2.542, time 6.99s, 
epoch 30, batches 2000|18183, train-acc 0.774, loss 0.555, para-norm 4742.020, grad-norm 3.246, time 6.96s, 
epoch 30, batches 3000|18183, train-acc 0.774, loss 0.555, para-norm 4747.001, grad-norm 4.916, time 6.95s, 
epoch 30, batches 4000|18183, train-acc 0.772, loss 0.556, para-norm 4751.846, grad-norm 1.050, time 6.95s, 
epoch 30, batches 5000|18183, train-acc 0.777, loss 0.546, para-norm 4754.259, grad-norm 2.382, time 6.94s, 
epoch 30, batches 6000|18183, train-acc 0.780, loss 0.545, para-norm 4752.216, grad-norm 1.593, time 6.94s, 
epoch 30, batches 7000|18183, train-acc 0.784, loss 0.541, para-norm 4750.855, grad-norm 2.501, time 6.94s, 
epoch 30, batches 8000|18183, train-acc 0.780, loss 0.546, para-norm 4749.317, grad-norm 4.732, time 6.95s, 
epoch 30, batches 9000|18183, train-acc 0.780, loss 0.545, para-norm 4748.769, grad-norm 3.003, time 6.95s, 
epoch 30, batches 10000|18183, train-acc 0.777, loss 0.553, para-norm 4750.346, grad-norm 4.047, time 6.96s, 
epoch 30, batches 11000|18183, train-acc 0.776, loss 0.543, para-norm 4749.726, grad-norm 0.975, time 6.94s, 
epoch 30, batches 12000|18183, train-acc 0.775, loss 0.551, para-norm 4748.747, grad-norm 2.238, time 6.95s, 
epoch 30, batches 13000|18183, train-acc 0.779, loss 0.544, para-norm 4748.960, grad-norm 3.116, time 6.94s, 
epoch 30, batches 14000|18183, train-acc 0.780, loss 0.544, para-norm 4747.434, grad-norm 1.385, time 6.95s, 
epoch 30, batches 15000|18183, train-acc 0.775, loss 0.556, para-norm 4745.759, grad-norm 2.422, time 6.95s, 
epoch 30, batches 16000|18183, train-acc 0.778, loss 0.547, para-norm 4745.768, grad-norm 3.317, time 6.97s, 
epoch 30, batches 17000|18183, train-acc 0.772, loss 0.559, para-norm 4744.550, grad-norm 2.949, time 6.95s, 
epoch 30, batches 18000|18183, train-acc 0.777, loss 0.550, para-norm 4744.514, grad-norm 1.147, time 6.94s, 
epoch 30, batches 18183|18183, train-acc 0.780, loss 0.543, para-norm 4744.877, grad-norm 4.083, time 1.27s, 
dev-acc 0.802
epoch 31, batches 1000|18183, train-acc 0.777, loss 0.549, para-norm 4749.602, grad-norm 2.611, time 7.00s, 
epoch 31, batches 2000|18183, train-acc 0.786, loss 0.537, para-norm 4747.982, grad-norm 2.856, time 6.95s, 
epoch 31, batches 3000|18183, train-acc 0.786, loss 0.536, para-norm 4757.689, grad-norm 2.936, time 6.94s, 
epoch 31, batches 4000|18183, train-acc 0.778, loss 0.549, para-norm 4757.798, grad-norm 3.102, time 6.95s, 
epoch 31, batches 5000|18183, train-acc 0.774, loss 0.550, para-norm 4754.574, grad-norm 3.019, time 6.96s, 
epoch 31, batches 6000|18183, train-acc 0.775, loss 0.553, para-norm 4758.395, grad-norm 2.830, time 7.03s, 
epoch 31, batches 7000|18183, train-acc 0.780, loss 0.547, para-norm 4759.079, grad-norm 2.418, time 6.97s, 
epoch 31, batches 8000|18183, train-acc 0.774, loss 0.556, para-norm 4758.181, grad-norm 2.438, time 6.96s, 
epoch 31, batches 9000|18183, train-acc 0.779, loss 0.543, para-norm 4756.780, grad-norm 3.244, time 6.95s, 
epoch 31, batches 10000|18183, train-acc 0.780, loss 0.544, para-norm 4776.287, grad-norm 4.798, time 6.95s, 
epoch 31, batches 11000|18183, train-acc 0.778, loss 0.547, para-norm 4774.859, grad-norm 8.462, time 6.96s, 
epoch 31, batches 12000|18183, train-acc 0.778, loss 0.549, para-norm 4773.037, grad-norm 2.719, time 6.96s, 
epoch 31, batches 13000|18183, train-acc 0.780, loss 0.544, para-norm 4770.774, grad-norm 3.826, time 6.96s, 
epoch 31, batches 14000|18183, train-acc 0.778, loss 0.550, para-norm 4769.433, grad-norm 2.293, time 6.96s, 
epoch 31, batches 15000|18183, train-acc 0.780, loss 0.546, para-norm 4766.819, grad-norm 1.805, time 6.96s, 
epoch 31, batches 16000|18183, train-acc 0.774, loss 0.552, para-norm 4766.120, grad-norm 2.938, time 6.94s, 
epoch 31, batches 17000|18183, train-acc 0.780, loss 0.545, para-norm 4764.390, grad-norm 2.251, time 6.95s, 
epoch 31, batches 18000|18183, train-acc 0.779, loss 0.545, para-norm 4763.103, grad-norm 1.933, time 6.95s, 
epoch 31, batches 18183|18183, train-acc 0.780, loss 0.555, para-norm 4762.771, grad-norm 2.534, time 1.28s, 
dev-acc 0.803
epoch 32, batches 1000|18183, train-acc 0.779, loss 0.545, para-norm 4764.331, grad-norm 2.206, time 6.99s, 
epoch 32, batches 2000|18183, train-acc 0.783, loss 0.541, para-norm 4764.369, grad-norm 1.302, time 6.96s, 
epoch 32, batches 3000|18183, train-acc 0.778, loss 0.546, para-norm 4762.900, grad-norm 3.270, time 6.95s, 
epoch 32, batches 4000|18183, train-acc 0.784, loss 0.536, para-norm 4768.631, grad-norm 2.972, time 6.96s, 
epoch 32, batches 5000|18183, train-acc 0.775, loss 0.552, para-norm 4767.598, grad-norm 1.984, time 6.98s, 
epoch 32, batches 6000|18183, train-acc 0.779, loss 0.542, para-norm 4767.392, grad-norm 57.026, time 6.96s, 
epoch 32, batches 7000|18183, train-acc 0.781, loss 0.540, para-norm 4767.699, grad-norm 3.256, time 6.94s, 
epoch 32, batches 8000|18183, train-acc 0.781, loss 0.545, para-norm 4765.985, grad-norm 1.993, time 6.95s, 
epoch 32, batches 9000|18183, train-acc 0.782, loss 0.548, para-norm 4763.736, grad-norm 2.361, time 6.95s, 
epoch 32, batches 10000|18183, train-acc 0.780, loss 0.546, para-norm 4761.907, grad-norm 5.549, time 6.96s, 
epoch 32, batches 11000|18183, train-acc 0.778, loss 0.547, para-norm 4760.518, grad-norm 2.981, time 6.94s, 
epoch 32, batches 12000|18183, train-acc 0.775, loss 0.552, para-norm 4758.681, grad-norm 3.136, time 6.95s, 
epoch 32, batches 13000|18183, train-acc 0.779, loss 0.547, para-norm 4758.080, grad-norm 0.763, time 6.96s, 
epoch 32, batches 14000|18183, train-acc 0.780, loss 0.548, para-norm 4757.398, grad-norm 1.436, time 6.95s, 
epoch 32, batches 15000|18183, train-acc 0.778, loss 0.549, para-norm 4756.038, grad-norm 7.139, time 6.95s, 
epoch 32, batches 16000|18183, train-acc 0.778, loss 0.549, para-norm 4755.517, grad-norm 9.445, time 6.95s, 
epoch 32, batches 17000|18183, train-acc 0.780, loss 0.542, para-norm 4756.198, grad-norm 2.382, time 6.95s, 
epoch 32, batches 18000|18183, train-acc 0.779, loss 0.542, para-norm 4754.261, grad-norm 3.882, time 7.03s, 
epoch 32, batches 18183|18183, train-acc 0.789, loss 0.518, para-norm 4754.466, grad-norm 3.037, time 1.26s, 
dev-acc 0.803
epoch 33, batches 1000|18183, train-acc 0.781, loss 0.539, para-norm 4753.654, grad-norm 2.143, time 7.04s, 
epoch 33, batches 2000|18183, train-acc 0.779, loss 0.546, para-norm 4751.463, grad-norm 2.201, time 6.94s, 
epoch 33, batches 3000|18183, train-acc 0.779, loss 0.545, para-norm 4752.360, grad-norm 5.044, time 6.95s, 
epoch 33, batches 4000|18183, train-acc 0.783, loss 0.540, para-norm 4761.816, grad-norm 2.902, time 6.95s, 
epoch 33, batches 5000|18183, train-acc 0.784, loss 0.539, para-norm 4763.864, grad-norm 5.141, time 6.93s, 
epoch 33, batches 6000|18183, train-acc 0.781, loss 0.545, para-norm 4762.771, grad-norm 9.094, time 6.96s, 
epoch 33, batches 7000|18183, train-acc 0.779, loss 0.544, para-norm 4762.201, grad-norm 15.544, time 6.97s, 
epoch 33, batches 8000|18183, train-acc 0.781, loss 0.544, para-norm 4766.473, grad-norm 3.074, time 6.95s, 
epoch 33, batches 9000|18183, train-acc 0.785, loss 0.540, para-norm 4765.199, grad-norm 0.900, time 6.94s, 
epoch 33, batches 10000|18183, train-acc 0.780, loss 0.545, para-norm 4762.908, grad-norm 4.986, time 6.94s, 
epoch 33, batches 11000|18183, train-acc 0.780, loss 0.543, para-norm 4762.115, grad-norm 4.316, time 6.94s, 
epoch 33, batches 12000|18183, train-acc 0.780, loss 0.542, para-norm 4760.009, grad-norm 126.698, time 7.24s, 
epoch 33, batches 13000|18183, train-acc 0.780, loss 0.545, para-norm 4759.930, grad-norm 1.439, time 7.21s, 
epoch 33, batches 14000|18183, train-acc 0.778, loss 0.550, para-norm 4759.202, grad-norm 3.159, time 7.22s, 
epoch 33, batches 15000|18183, train-acc 0.782, loss 0.539, para-norm 4759.656, grad-norm 2.812, time 7.20s, 
epoch 33, batches 16000|18183, train-acc 0.781, loss 0.543, para-norm 4760.098, grad-norm 2.044, time 7.23s, 
epoch 33, batches 17000|18183, train-acc 0.786, loss 0.534, para-norm 4757.394, grad-norm 3.398, time 7.22s, 
epoch 33, batches 18000|18183, train-acc 0.780, loss 0.543, para-norm 4757.795, grad-norm 3.240, time 7.20s, 
epoch 33, batches 18183|18183, train-acc 0.784, loss 0.547, para-norm 4757.448, grad-norm 18.686, time 1.32s, 
dev-acc 0.807
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
save model!
epoch 34, batches 1000|18183, train-acc 0.786, loss 0.536, para-norm 4756.818, grad-norm 2.399, time 7.31s, 
epoch 34, batches 2000|18183, train-acc 0.784, loss 0.536, para-norm 4755.241, grad-norm 2.670, time 7.20s, 
epoch 34, batches 3000|18183, train-acc 0.780, loss 0.538, para-norm 4755.236, grad-norm 23.293, time 7.21s, 
epoch 34, batches 4000|18183, train-acc 0.784, loss 0.532, para-norm 4755.643, grad-norm 5.246, time 7.20s, 
epoch 34, batches 5000|18183, train-acc 0.785, loss 0.537, para-norm 4754.603, grad-norm 2.788, time 7.21s, 
epoch 34, batches 6000|18183, train-acc 0.780, loss 0.546, para-norm 4754.944, grad-norm 2.935, time 7.22s, 
epoch 34, batches 7000|18183, train-acc 0.780, loss 0.548, para-norm 4755.521, grad-norm 19.617, time 7.20s, 
epoch 34, batches 8000|18183, train-acc 0.782, loss 0.542, para-norm 4753.047, grad-norm 2.839, time 7.20s, 
epoch 34, batches 9000|18183, train-acc 0.785, loss 0.535, para-norm 4753.590, grad-norm 2.030, time 7.19s, 
epoch 34, batches 10000|18183, train-acc 0.782, loss 0.541, para-norm 4754.397, grad-norm 57.583, time 7.20s, 
epoch 34, batches 11000|18183, train-acc 0.778, loss 0.549, para-norm 4751.865, grad-norm 52.368, time 7.20s, 
epoch 34, batches 12000|18183, train-acc 0.782, loss 0.540, para-norm 4754.732, grad-norm 6.815, time 7.20s, 
epoch 34, batches 13000|18183, train-acc 0.783, loss 0.533, para-norm 4754.167, grad-norm 3.738, time 7.20s, 
epoch 34, batches 14000|18183, train-acc 0.779, loss 0.547, para-norm 4753.229, grad-norm 3.674, time 7.21s, 
epoch 34, batches 15000|18183, train-acc 0.781, loss 0.545, para-norm 4758.430, grad-norm 2.442, time 7.18s, 
epoch 34, batches 16000|18183, train-acc 0.784, loss 0.538, para-norm 4764.220, grad-norm 12.747, time 7.20s, 
epoch 34, batches 17000|18183, train-acc 0.787, loss 0.532, para-norm 4760.029, grad-norm 2.290, time 7.20s, 
epoch 34, batches 18000|18183, train-acc 0.782, loss 0.538, para-norm 4758.465, grad-norm 40.835, time 7.27s, 
epoch 34, batches 18183|18183, train-acc 0.782, loss 0.542, para-norm 4759.088, grad-norm 7.511, time 1.32s, 
dev-acc 0.805
epoch 35, batches 1000|18183, train-acc 0.786, loss 0.532, para-norm 4760.358, grad-norm 2.115, time 7.26s, 
epoch 35, batches 2000|18183, train-acc 0.785, loss 0.532, para-norm 4760.879, grad-norm 2.845, time 7.20s, 
epoch 35, batches 3000|18183, train-acc 0.786, loss 0.534, para-norm 4757.994, grad-norm 1.936, time 7.20s, 
epoch 35, batches 4000|18183, train-acc 0.785, loss 0.533, para-norm 4757.182, grad-norm 4.688, time 7.20s, 
epoch 35, batches 5000|18183, train-acc 0.781, loss 0.541, para-norm 4755.434, grad-norm 12.378, time 7.20s, 
epoch 35, batches 6000|18183, train-acc 0.782, loss 0.538, para-norm 4756.903, grad-norm 3.931, time 7.20s, 
epoch 35, batches 7000|18183, train-acc 0.786, loss 0.529, para-norm 4757.451, grad-norm 12.250, time 7.21s, 
epoch 35, batches 8000|18183, train-acc 0.779, loss 0.544, para-norm 4756.856, grad-norm 4.383, time 7.20s, 
epoch 35, batches 9000|18183, train-acc 0.783, loss 0.537, para-norm 4756.928, grad-norm 4.378, time 7.20s, 
epoch 35, batches 10000|18183, train-acc 0.779, loss 0.543, para-norm 4755.996, grad-norm 4.110, time 7.21s, 
epoch 35, batches 11000|18183, train-acc 0.785, loss 0.536, para-norm 4755.603, grad-norm 1.995, time 7.19s, 
epoch 35, batches 12000|18183, train-acc 0.786, loss 0.536, para-norm 4759.282, grad-norm 7.299, time 7.20s, 
epoch 35, batches 13000|18183, train-acc 0.783, loss 0.540, para-norm 4760.598, grad-norm 3.056, time 7.19s, 
epoch 35, batches 14000|18183, train-acc 0.784, loss 0.539, para-norm 4759.489, grad-norm 2.928, time 7.20s, 
epoch 35, batches 15000|18183, train-acc 0.782, loss 0.540, para-norm 4758.753, grad-norm 2.579, time 7.21s, 
epoch 35, batches 16000|18183, train-acc 0.781, loss 0.545, para-norm 4758.900, grad-norm 2.671, time 7.20s, 
epoch 35, batches 17000|18183, train-acc 0.783, loss 0.536, para-norm 4765.706, grad-norm 2.937, time 7.19s, 
epoch 35, batches 18000|18183, train-acc 0.786, loss 0.535, para-norm 4764.594, grad-norm 3.373, time 7.20s, 
epoch 35, batches 18183|18183, train-acc 0.779, loss 0.553, para-norm 4764.519, grad-norm 2.047, time 1.32s, 
dev-acc 0.812
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
save model!
epoch 36, batches 1000|18183, train-acc 0.781, loss 0.544, para-norm 4765.015, grad-norm 3.333, time 7.16s, 
epoch 36, batches 2000|18183, train-acc 0.786, loss 0.530, para-norm 4767.379, grad-norm 3.993, time 6.94s, 
epoch 36, batches 3000|18183, train-acc 0.781, loss 0.544, para-norm 4767.177, grad-norm 5.280, time 6.94s, 
epoch 36, batches 4000|18183, train-acc 0.786, loss 0.533, para-norm 4766.942, grad-norm 2.659, time 6.94s, 
epoch 36, batches 5000|18183, train-acc 0.782, loss 0.539, para-norm 4768.130, grad-norm 2.890, time 6.94s, 
epoch 36, batches 6000|18183, train-acc 0.782, loss 0.536, para-norm 4768.217, grad-norm 1.663, time 6.94s, 
epoch 36, batches 7000|18183, train-acc 0.788, loss 0.530, para-norm 4768.802, grad-norm 2.141, time 6.98s, 
epoch 36, batches 8000|18183, train-acc 0.782, loss 0.542, para-norm 4767.155, grad-norm 3.148, time 6.94s, 
epoch 36, batches 9000|18183, train-acc 0.785, loss 0.531, para-norm 4766.537, grad-norm 2.913, time 6.94s, 
epoch 36, batches 10000|18183, train-acc 0.786, loss 0.530, para-norm 4764.683, grad-norm 1.506, time 7.78s, 
epoch 36, batches 11000|18183, train-acc 0.785, loss 0.536, para-norm 4763.547, grad-norm 3.177, time 7.77s, 
epoch 36, batches 12000|18183, train-acc 0.790, loss 0.526, para-norm 4764.078, grad-norm 1.452, time 7.75s, 
epoch 36, batches 13000|18183, train-acc 0.783, loss 0.539, para-norm 4763.175, grad-norm 2.153, time 7.77s, 
epoch 36, batches 14000|18183, train-acc 0.784, loss 0.533, para-norm 4762.053, grad-norm 3.895, time 7.77s, 
epoch 36, batches 15000|18183, train-acc 0.782, loss 0.541, para-norm 4760.690, grad-norm 3.221, time 7.74s, 
epoch 36, batches 16000|18183, train-acc 0.779, loss 0.545, para-norm 4760.093, grad-norm 47.856, time 7.75s, 
epoch 36, batches 17000|18183, train-acc 0.787, loss 0.532, para-norm 4757.726, grad-norm 3.805, time 7.80s, 
epoch 36, batches 18000|18183, train-acc 0.788, loss 0.529, para-norm 4757.561, grad-norm 4.965, time 7.75s, 
epoch 36, batches 18183|18183, train-acc 0.796, loss 0.520, para-norm 4757.722, grad-norm 2.937, time 1.42s, 
dev-acc 0.808
epoch 37, batches 1000|18183, train-acc 0.789, loss 0.525, para-norm 4757.744, grad-norm 12.420, time 7.82s, 
epoch 37, batches 2000|18183, train-acc 0.784, loss 0.536, para-norm 4757.817, grad-norm 2.187, time 7.78s, 
epoch 37, batches 3000|18183, train-acc 0.787, loss 0.532, para-norm 4757.335, grad-norm 2.420, time 7.76s, 
epoch 37, batches 4000|18183, train-acc 0.789, loss 0.521, para-norm 4757.153, grad-norm 1.822, time 7.75s, 
epoch 37, batches 5000|18183, train-acc 0.791, loss 0.522, para-norm 4764.580, grad-norm 4.870, time 7.75s, 
epoch 37, batches 6000|18183, train-acc 0.784, loss 0.535, para-norm 4763.915, grad-norm 6.367, time 7.78s, 
epoch 37, batches 7000|18183, train-acc 0.783, loss 0.535, para-norm 4765.781, grad-norm 1.091, time 7.75s, 
epoch 37, batches 8000|18183, train-acc 0.786, loss 0.535, para-norm 4765.969, grad-norm 3.752, time 7.76s, 
epoch 37, batches 9000|18183, train-acc 0.788, loss 0.529, para-norm 4765.420, grad-norm 4.545, time 7.76s, 
epoch 37, batches 10000|18183, train-acc 0.785, loss 0.535, para-norm 4764.616, grad-norm 2.020, time 7.77s, 
epoch 37, batches 11000|18183, train-acc 0.783, loss 0.534, para-norm 4769.349, grad-norm 1.603, time 7.75s, 
epoch 37, batches 12000|18183, train-acc 0.780, loss 0.541, para-norm 4770.985, grad-norm 3.019, time 7.75s, 
epoch 37, batches 13000|18183, train-acc 0.786, loss 0.534, para-norm 4771.933, grad-norm 1.200, time 7.77s, 
epoch 37, batches 14000|18183, train-acc 0.783, loss 0.534, para-norm 4770.536, grad-norm 3.842, time 7.76s, 
epoch 37, batches 15000|18183, train-acc 0.784, loss 0.536, para-norm 4769.869, grad-norm 1.469, time 7.78s, 
epoch 37, batches 16000|18183, train-acc 0.783, loss 0.536, para-norm 4769.396, grad-norm 3.240, time 7.78s, 
epoch 37, batches 17000|18183, train-acc 0.788, loss 0.532, para-norm 4769.582, grad-norm 3.388, time 7.76s, 
epoch 37, batches 18000|18183, train-acc 0.781, loss 0.546, para-norm 4767.808, grad-norm 2.569, time 7.76s, 
epoch 37, batches 18183|18183, train-acc 0.785, loss 0.532, para-norm 4767.389, grad-norm 4.630, time 1.42s, 
dev-acc 0.806
epoch 38, batches 1000|18183, train-acc 0.789, loss 0.524, para-norm 4767.735, grad-norm 3.669, time 7.88s, 
epoch 38, batches 2000|18183, train-acc 0.791, loss 0.521, para-norm 4770.669, grad-norm 3.375, time 6.95s, 
epoch 38, batches 3000|18183, train-acc 0.786, loss 0.530, para-norm 4778.958, grad-norm 1.958, time 6.93s, 
epoch 38, batches 4000|18183, train-acc 0.787, loss 0.532, para-norm 4777.698, grad-norm 35.923, time 6.93s, 
epoch 38, batches 5000|18183, train-acc 0.787, loss 0.534, para-norm 4778.592, grad-norm 4.207, time 6.94s, 
epoch 38, batches 6000|18183, train-acc 0.786, loss 0.534, para-norm 4778.664, grad-norm 5.227, time 6.95s, 
epoch 38, batches 7000|18183, train-acc 0.789, loss 0.523, para-norm 4778.024, grad-norm 2.076, time 6.94s, 
epoch 38, batches 8000|18183, train-acc 0.784, loss 0.530, para-norm 4776.215, grad-norm 2.974, time 6.94s, 
epoch 38, batches 9000|18183, train-acc 0.789, loss 0.529, para-norm 4774.624, grad-norm 1.514, time 6.95s, 
epoch 38, batches 10000|18183, train-acc 0.792, loss 0.528, para-norm 4775.791, grad-norm 2.255, time 6.95s, 
epoch 38, batches 11000|18183, train-acc 0.787, loss 0.533, para-norm 4775.571, grad-norm 3.384, time 6.94s, 
epoch 38, batches 12000|18183, train-acc 0.787, loss 0.529, para-norm 4774.281, grad-norm 6.031, time 6.96s, 
epoch 38, batches 13000|18183, train-acc 0.785, loss 0.529, para-norm 4772.151, grad-norm 3.284, time 6.93s, 
epoch 38, batches 14000|18183, train-acc 0.785, loss 0.532, para-norm 4770.398, grad-norm 2.199, time 6.94s, 
epoch 38, batches 15000|18183, train-acc 0.785, loss 0.530, para-norm 4769.804, grad-norm 2.054, time 6.95s, 
epoch 38, batches 16000|18183, train-acc 0.787, loss 0.530, para-norm 4770.058, grad-norm 0.864, time 6.95s, 
epoch 38, batches 17000|18183, train-acc 0.787, loss 0.529, para-norm 4770.067, grad-norm 3.052, time 6.94s, 
epoch 38, batches 18000|18183, train-acc 0.780, loss 0.540, para-norm 4769.246, grad-norm 1.328, time 6.96s, 
epoch 38, batches 18183|18183, train-acc 0.787, loss 0.528, para-norm 4769.004, grad-norm 19.394, time 1.27s, 
dev-acc 0.810
epoch 39, batches 1000|18183, train-acc 0.788, loss 0.524, para-norm 4770.314, grad-norm 2.476, time 7.05s, 
epoch 39, batches 2000|18183, train-acc 0.785, loss 0.533, para-norm 4771.732, grad-norm 2.262, time 6.94s, 
epoch 39, batches 3000|18183, train-acc 0.788, loss 0.526, para-norm 4773.062, grad-norm 3.019, time 6.94s, 
epoch 39, batches 4000|18183, train-acc 0.783, loss 0.532, para-norm 4771.439, grad-norm 2.896, time 6.94s, 
epoch 39, batches 5000|18183, train-acc 0.785, loss 0.529, para-norm 4771.980, grad-norm 3.023, time 6.95s, 
epoch 39, batches 6000|18183, train-acc 0.787, loss 0.528, para-norm 4771.491, grad-norm 3.567, time 6.95s, 
epoch 39, batches 7000|18183, train-acc 0.789, loss 0.524, para-norm 4770.100, grad-norm 2.175, time 6.95s, 
epoch 39, batches 8000|18183, train-acc 0.784, loss 0.536, para-norm 4769.310, grad-norm 2.496, time 6.95s, 
epoch 39, batches 9000|18183, train-acc 0.789, loss 0.530, para-norm 4767.105, grad-norm 2.667, time 6.93s, 
epoch 39, batches 10000|18183, train-acc 0.788, loss 0.528, para-norm 4766.368, grad-norm 2.280, time 6.95s, 
epoch 39, batches 11000|18183, train-acc 0.789, loss 0.528, para-norm 4767.242, grad-norm 3.195, time 6.96s, 
epoch 39, batches 12000|18183, train-acc 0.789, loss 0.530, para-norm 4766.590, grad-norm 3.402, time 6.94s, 
epoch 39, batches 13000|18183, train-acc 0.786, loss 0.527, para-norm 4767.223, grad-norm 3.635, time 6.93s, 
epoch 39, batches 14000|18183, train-acc 0.788, loss 0.532, para-norm 4766.648, grad-norm 2.552, time 6.95s, 
epoch 39, batches 15000|18183, train-acc 0.790, loss 0.522, para-norm 4766.782, grad-norm 9.070, time 6.94s, 
epoch 39, batches 16000|18183, train-acc 0.792, loss 0.523, para-norm 4765.950, grad-norm 3.489, time 6.95s, 
epoch 39, batches 17000|18183, train-acc 0.788, loss 0.526, para-norm 4765.925, grad-norm 15.956, time 6.94s, 
epoch 39, batches 18000|18183, train-acc 0.788, loss 0.526, para-norm 4764.440, grad-norm 16.781, time 6.95s, 
epoch 39, batches 18183|18183, train-acc 0.782, loss 0.543, para-norm 4764.655, grad-norm 2.187, time 1.28s, 
dev-acc 0.810
epoch 40, batches 1000|18183, train-acc 0.790, loss 0.526, para-norm 4765.442, grad-norm 5.304, time 6.99s, 
epoch 40, batches 2000|18183, train-acc 0.787, loss 0.531, para-norm 4765.063, grad-norm 1.195, time 6.96s, 
epoch 40, batches 3000|18183, train-acc 0.786, loss 0.529, para-norm 4764.483, grad-norm 5.055, time 6.95s, 
epoch 40, batches 4000|18183, train-acc 0.790, loss 0.522, para-norm 4765.134, grad-norm 2.521, time 6.95s, 
epoch 40, batches 5000|18183, train-acc 0.790, loss 0.523, para-norm 4764.913, grad-norm 11.334, time 7.01s, 
epoch 40, batches 6000|18183, train-acc 0.791, loss 0.522, para-norm 4766.443, grad-norm 2.931, time 6.95s, 
epoch 40, batches 7000|18183, train-acc 0.784, loss 0.530, para-norm 4765.204, grad-norm 2.758, time 6.95s, 
epoch 40, batches 8000|18183, train-acc 0.791, loss 0.527, para-norm 4763.600, grad-norm 48.542, time 6.95s, 
epoch 40, batches 9000|18183, train-acc 0.789, loss 0.524, para-norm 4764.654, grad-norm 8.260, time 6.94s, 
epoch 40, batches 10000|18183, train-acc 0.787, loss 0.527, para-norm 4763.163, grad-norm 3.623, time 6.96s, 
epoch 40, batches 11000|18183, train-acc 0.787, loss 0.533, para-norm 4762.081, grad-norm 76.647, time 6.94s, 
epoch 40, batches 12000|18183, train-acc 0.789, loss 0.528, para-norm 4767.168, grad-norm 256.511, time 6.97s, 
epoch 40, batches 13000|18183, train-acc 0.788, loss 0.526, para-norm 4769.777, grad-norm 3.816, time 6.96s, 
epoch 40, batches 14000|18183, train-acc 0.789, loss 0.527, para-norm 4769.552, grad-norm 2.057, time 6.94s, 
epoch 40, batches 15000|18183, train-acc 0.785, loss 0.533, para-norm 4769.877, grad-norm 3.012, time 6.97s, 
epoch 40, batches 16000|18183, train-acc 0.789, loss 0.523, para-norm 4769.738, grad-norm 6.184, time 7.08s, 
epoch 40, batches 17000|18183, train-acc 0.791, loss 0.521, para-norm 4768.311, grad-norm 3.835, time 6.94s, 
epoch 40, batches 18000|18183, train-acc 0.789, loss 0.528, para-norm 4773.385, grad-norm 4.086, time 6.95s, 
epoch 40, batches 18183|18183, train-acc 0.797, loss 0.519, para-norm 4774.684, grad-norm 0.833, time 1.27s, 
dev-acc 0.812
epoch 41, batches 1000|18183, train-acc 0.785, loss 0.529, para-norm 4773.556, grad-norm 6.414, time 7.04s, 
epoch 41, batches 2000|18183, train-acc 0.792, loss 0.518, para-norm 4786.099, grad-norm 6.787, time 6.95s, 
epoch 41, batches 3000|18183, train-acc 0.790, loss 0.522, para-norm 4791.921, grad-norm 61.907, time 6.96s, 
epoch 41, batches 4000|18183, train-acc 0.787, loss 0.531, para-norm 4793.439, grad-norm 2.687, time 6.94s, 
epoch 41, batches 5000|18183, train-acc 0.791, loss 0.521, para-norm 4794.107, grad-norm 1.902, time 7.06s, 
epoch 41, batches 6000|18183, train-acc 0.791, loss 0.525, para-norm 4791.317, grad-norm 1.262, time 6.95s, 
epoch 41, batches 7000|18183, train-acc 0.791, loss 0.523, para-norm 4791.194, grad-norm 3.294, time 6.95s, 
epoch 41, batches 8000|18183, train-acc 0.788, loss 0.529, para-norm 4788.387, grad-norm 2.549, time 6.97s, 
epoch 41, batches 9000|18183, train-acc 0.785, loss 0.529, para-norm 4789.634, grad-norm 2.523, time 6.97s, 
epoch 41, batches 10000|18183, train-acc 0.787, loss 0.526, para-norm 4789.308, grad-norm 6.238, time 6.96s, 
epoch 41, batches 11000|18183, train-acc 0.790, loss 0.522, para-norm 4788.512, grad-norm 144.409, time 6.93s, 
epoch 41, batches 12000|18183, train-acc 0.793, loss 0.511, para-norm 4787.790, grad-norm 2.926, time 6.95s, 
epoch 41, batches 13000|18183, train-acc 0.790, loss 0.526, para-norm 4786.516, grad-norm 6.759, time 6.94s, 
epoch 41, batches 14000|18183, train-acc 0.787, loss 0.524, para-norm 4786.980, grad-norm 6.981, time 6.95s, 
epoch 41, batches 15000|18183, train-acc 0.790, loss 0.526, para-norm 4784.155, grad-norm 46.770, time 6.95s, 
epoch 41, batches 16000|18183, train-acc 0.783, loss 0.541, para-norm 4782.525, grad-norm 5.449, time 6.96s, 
epoch 41, batches 17000|18183, train-acc 0.791, loss 0.520, para-norm 4781.171, grad-norm 5.772, time 6.96s, 
epoch 41, batches 18000|18183, train-acc 0.791, loss 0.521, para-norm 4784.762, grad-norm 3.947, time 7.03s, 
epoch 41, batches 18183|18183, train-acc 0.787, loss 0.525, para-norm 4785.226, grad-norm 3.579, time 1.27s, 
dev-acc 0.814
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
save model!
epoch 42, batches 1000|18183, train-acc 0.787, loss 0.528, para-norm 4789.643, grad-norm 4.533, time 7.11s, 
epoch 42, batches 2000|18183, train-acc 0.790, loss 0.520, para-norm 4786.566, grad-norm 3.426, time 6.93s, 
epoch 42, batches 3000|18183, train-acc 0.788, loss 0.521, para-norm 4785.773, grad-norm 6.305, time 6.94s, 
epoch 42, batches 4000|18183, train-acc 0.789, loss 0.515, para-norm 4794.517, grad-norm 3.204, time 6.94s, 
epoch 42, batches 5000|18183, train-acc 0.792, loss 0.519, para-norm 4791.999, grad-norm 4.636, time 6.95s, 
epoch 42, batches 6000|18183, train-acc 0.787, loss 0.529, para-norm 4792.179, grad-norm 4.696, time 6.95s, 
epoch 42, batches 7000|18183, train-acc 0.786, loss 0.530, para-norm 4791.236, grad-norm 4.524, time 6.95s, 
epoch 42, batches 8000|18183, train-acc 0.788, loss 0.523, para-norm 4800.470, grad-norm 4.253, time 6.95s, 
epoch 42, batches 9000|18183, train-acc 0.792, loss 0.517, para-norm 4802.896, grad-norm 2.756, time 6.94s, 
epoch 42, batches 10000|18183, train-acc 0.791, loss 0.523, para-norm 4802.119, grad-norm 4.015, time 6.93s, 
epoch 42, batches 11000|18183, train-acc 0.793, loss 0.522, para-norm 4806.770, grad-norm 1.107, time 6.94s, 
epoch 42, batches 12000|18183, train-acc 0.788, loss 0.528, para-norm 4807.620, grad-norm 1.872, time 6.96s, 
epoch 42, batches 13000|18183, train-acc 0.789, loss 0.526, para-norm 4807.506, grad-norm 0.933, time 6.95s, 
epoch 42, batches 14000|18183, train-acc 0.790, loss 0.521, para-norm 4806.926, grad-norm 4.678, time 6.95s, 
epoch 42, batches 15000|18183, train-acc 0.790, loss 0.523, para-norm 4806.032, grad-norm 7.435, time 6.96s, 
epoch 42, batches 16000|18183, train-acc 0.787, loss 0.522, para-norm 4805.512, grad-norm 1.709, time 6.95s, 
epoch 42, batches 17000|18183, train-acc 0.787, loss 0.527, para-norm 4803.398, grad-norm 3.933, time 6.96s, 
epoch 42, batches 18000|18183, train-acc 0.792, loss 0.519, para-norm 4801.310, grad-norm 2.667, time 6.94s, 
epoch 42, batches 18183|18183, train-acc 0.793, loss 0.515, para-norm 4801.267, grad-norm 4.263, time 1.27s, 
dev-acc 0.812
epoch 43, batches 1000|18183, train-acc 0.789, loss 0.522, para-norm 4801.634, grad-norm 1.917, time 7.03s, 
epoch 43, batches 2000|18183, train-acc 0.794, loss 0.515, para-norm 4801.489, grad-norm 3.932, time 6.97s, 
epoch 43, batches 3000|18183, train-acc 0.785, loss 0.530, para-norm 4797.974, grad-norm 3.379, time 6.96s, 
epoch 43, batches 4000|18183, train-acc 0.795, loss 0.515, para-norm 4796.950, grad-norm 10.317, time 6.94s, 
epoch 43, batches 5000|18183, train-acc 0.789, loss 0.524, para-norm 4804.626, grad-norm 2.566, time 6.96s, 
epoch 43, batches 6000|18183, train-acc 0.791, loss 0.526, para-norm 4809.311, grad-norm 3.718, time 6.95s, 
epoch 43, batches 7000|18183, train-acc 0.787, loss 0.522, para-norm 4807.270, grad-norm 3.494, time 6.94s, 
epoch 43, batches 8000|18183, train-acc 0.796, loss 0.514, para-norm 4817.367, grad-norm 4.229, time 6.96s, 
epoch 43, batches 9000|18183, train-acc 0.791, loss 0.521, para-norm 4817.419, grad-norm 8.486, time 6.96s, 
epoch 43, batches 10000|18183, train-acc 0.791, loss 0.518, para-norm 4817.559, grad-norm 3.116, time 7.22s, 
epoch 43, batches 11000|18183, train-acc 0.788, loss 0.523, para-norm 4816.277, grad-norm 12.458, time 7.78s, 
epoch 43, batches 12000|18183, train-acc 0.791, loss 0.521, para-norm 4813.940, grad-norm 3.945, time 7.77s, 
epoch 43, batches 13000|18183, train-acc 0.795, loss 0.516, para-norm 4814.598, grad-norm 2.703, time 7.78s, 
epoch 43, batches 14000|18183, train-acc 0.789, loss 0.522, para-norm 4813.464, grad-norm 2.421, time 7.77s, 
epoch 43, batches 15000|18183, train-acc 0.795, loss 0.515, para-norm 4822.001, grad-norm 4.003, time 7.77s, 
epoch 43, batches 16000|18183, train-acc 0.786, loss 0.526, para-norm 4819.867, grad-norm 4.681, time 7.79s, 
epoch 43, batches 17000|18183, train-acc 0.792, loss 0.522, para-norm 4819.451, grad-norm 3.073, time 7.78s, 
epoch 43, batches 18000|18183, train-acc 0.794, loss 0.513, para-norm 4818.981, grad-norm 4.040, time 7.78s, 
epoch 43, batches 18183|18183, train-acc 0.787, loss 0.526, para-norm 4819.225, grad-norm 3.895, time 1.42s, 
dev-acc 0.815
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
save model!
epoch 44, batches 1000|18183, train-acc 0.791, loss 0.520, para-norm 4817.517, grad-norm 3.311, time 7.81s, 
epoch 44, batches 2000|18183, train-acc 0.794, loss 0.516, para-norm 4815.681, grad-norm 3.499, time 7.79s, 
epoch 44, batches 3000|18183, train-acc 0.798, loss 0.510, para-norm 4814.714, grad-norm 2.937, time 7.79s, 
epoch 44, batches 4000|18183, train-acc 0.796, loss 0.510, para-norm 4813.169, grad-norm 1.937, time 7.83s, 
epoch 44, batches 5000|18183, train-acc 0.790, loss 0.520, para-norm 4812.508, grad-norm 7.961, time 7.78s, 
epoch 44, batches 6000|18183, train-acc 0.791, loss 0.516, para-norm 4811.343, grad-norm 6.807, time 7.77s, 
epoch 44, batches 7000|18183, train-acc 0.792, loss 0.521, para-norm 4810.630, grad-norm 1.881, time 7.78s, 
epoch 44, batches 8000|18183, train-acc 0.790, loss 0.522, para-norm 4809.543, grad-norm 3.985, time 7.92s, 
epoch 44, batches 9000|18183, train-acc 0.793, loss 0.521, para-norm 4808.628, grad-norm 3.765, time 7.78s, 
epoch 44, batches 10000|18183, train-acc 0.794, loss 0.519, para-norm 4822.531, grad-norm 1.923, time 7.76s, 
epoch 44, batches 11000|18183, train-acc 0.792, loss 0.520, para-norm 4824.875, grad-norm 10.652, time 7.77s, 
epoch 44, batches 12000|18183, train-acc 0.793, loss 0.519, para-norm 4823.341, grad-norm 2.433, time 7.75s, 
epoch 44, batches 13000|18183, train-acc 0.794, loss 0.515, para-norm 4837.878, grad-norm 5.144, time 7.77s, 
epoch 44, batches 14000|18183, train-acc 0.791, loss 0.518, para-norm 4837.469, grad-norm 2.253, time 7.78s, 
epoch 44, batches 15000|18183, train-acc 0.795, loss 0.508, para-norm 4837.265, grad-norm 946.384, time 7.75s, 
epoch 44, batches 16000|18183, train-acc 0.790, loss 0.523, para-norm 4835.299, grad-norm 2.414, time 7.77s, 
epoch 44, batches 17000|18183, train-acc 0.784, loss 0.532, para-norm 4834.443, grad-norm 3.323, time 7.77s, 
epoch 44, batches 18000|18183, train-acc 0.792, loss 0.518, para-norm 4834.500, grad-norm 7.532, time 7.84s, 
epoch 44, batches 18183|18183, train-acc 0.800, loss 0.509, para-norm 4833.865, grad-norm 2.251, time 1.42s, 
dev-acc 0.812
epoch 45, batches 1000|18183, train-acc 0.794, loss 0.516, para-norm 4833.766, grad-norm 1.581, time 7.81s, 
epoch 45, batches 2000|18183, train-acc 0.788, loss 0.523, para-norm 4834.382, grad-norm 2.719, time 7.76s, 
epoch 45, batches 3000|18183, train-acc 0.791, loss 0.518, para-norm 4834.533, grad-norm 3.826, time 7.75s, 
epoch 45, batches 4000|18183, train-acc 0.793, loss 0.514, para-norm 4834.069, grad-norm 1.182, time 7.77s, 
epoch 45, batches 5000|18183, train-acc 0.789, loss 0.522, para-norm 4834.937, grad-norm 1.913, time 7.78s, 
epoch 45, batches 6000|18183, train-acc 0.792, loss 0.518, para-norm 4835.800, grad-norm 3.883, time 7.76s, 
epoch 45, batches 7000|18183, train-acc 0.792, loss 0.516, para-norm 4834.666, grad-norm 2.450, time 7.77s, 
epoch 45, batches 8000|18183, train-acc 0.793, loss 0.518, para-norm 4842.363, grad-norm 8.742, time 7.78s, 
epoch 45, batches 9000|18183, train-acc 0.798, loss 0.507, para-norm 4843.455, grad-norm 1.979, time 7.77s, 
epoch 45, batches 10000|18183, train-acc 0.797, loss 0.514, para-norm 4843.537, grad-norm 2.001, time 7.75s, 
epoch 45, batches 11000|18183, train-acc 0.797, loss 0.509, para-norm 4841.753, grad-norm 1.848, time 7.78s, 
epoch 45, batches 12000|18183, train-acc 0.796, loss 0.508, para-norm 4841.638, grad-norm 3.215, time 7.82s, 
epoch 45, batches 13000|18183, train-acc 0.792, loss 0.518, para-norm 4840.955, grad-norm 1.970, time 7.76s, 
epoch 45, batches 14000|18183, train-acc 0.794, loss 0.510, para-norm 4842.113, grad-norm 2.819, time 7.76s, 
epoch 45, batches 15000|18183, train-acc 0.791, loss 0.518, para-norm 4838.423, grad-norm 2.071, time 7.77s, 
epoch 45, batches 16000|18183, train-acc 0.795, loss 0.514, para-norm 4838.655, grad-norm 5.728, time 7.77s, 
epoch 45, batches 17000|18183, train-acc 0.791, loss 0.520, para-norm 4837.834, grad-norm 1.882, time 7.88s, 
epoch 45, batches 18000|18183, train-acc 0.788, loss 0.525, para-norm 4836.980, grad-norm 3.120, time 7.77s, 
epoch 45, batches 18183|18183, train-acc 0.791, loss 0.517, para-norm 4837.023, grad-norm 2.223, time 1.42s, 
dev-acc 0.818
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
save model!
epoch 46, batches 1000|18183, train-acc 0.793, loss 0.514, para-norm 4836.285, grad-norm 3.569, time 7.86s, 
epoch 46, batches 2000|18183, train-acc 0.798, loss 0.499, para-norm 4836.777, grad-norm 26.290, time 7.77s, 
epoch 46, batches 3000|18183, train-acc 0.793, loss 0.512, para-norm 4836.985, grad-norm 3.275, time 7.77s, 
epoch 46, batches 4000|18183, train-acc 0.790, loss 0.521, para-norm 4838.276, grad-norm 3.061, time 7.79s, 
epoch 46, batches 5000|18183, train-acc 0.795, loss 0.510, para-norm 4837.439, grad-norm 3.959, time 7.61s, 
epoch 46, batches 6000|18183, train-acc 0.789, loss 0.518, para-norm 4836.814, grad-norm 2.493, time 7.03s, 
epoch 46, batches 7000|18183, train-acc 0.791, loss 0.518, para-norm 4844.356, grad-norm 2.914, time 6.96s, 
epoch 46, batches 8000|18183, train-acc 0.798, loss 0.507, para-norm 4844.570, grad-norm 2.731, time 6.98s, 
epoch 46, batches 9000|18183, train-acc 0.791, loss 0.520, para-norm 4845.220, grad-norm 0.072, time 6.96s, 
epoch 46, batches 10000|18183, train-acc 0.791, loss 0.522, para-norm 4844.423, grad-norm 4.294, time 6.95s, 
epoch 46, batches 11000|18183, train-acc 0.794, loss 0.513, para-norm 4844.105, grad-norm 3.108, time 7.02s, 
epoch 46, batches 12000|18183, train-acc 0.791, loss 0.521, para-norm 4841.895, grad-norm 16.146, time 6.98s, 
epoch 46, batches 13000|18183, train-acc 0.794, loss 0.513, para-norm 4840.025, grad-norm 4.530, time 6.95s, 
epoch 46, batches 14000|18183, train-acc 0.796, loss 0.510, para-norm 4841.822, grad-norm 2.386, time 6.96s, 
epoch 46, batches 15000|18183, train-acc 0.789, loss 0.526, para-norm 4841.419, grad-norm 2.578, time 6.95s, 
epoch 46, batches 16000|18183, train-acc 0.798, loss 0.508, para-norm 4842.642, grad-norm 1.521, time 6.94s, 
epoch 46, batches 17000|18183, train-acc 0.796, loss 0.510, para-norm 4840.657, grad-norm 2.815, time 6.96s, 
epoch 46, batches 18000|18183, train-acc 0.794, loss 0.517, para-norm 4840.696, grad-norm 2.904, time 6.96s, 
epoch 46, batches 18183|18183, train-acc 0.796, loss 0.517, para-norm 4841.135, grad-norm 1.929, time 1.27s, 
dev-acc 0.816
epoch 47, batches 1000|18183, train-acc 0.795, loss 0.515, para-norm 4840.461, grad-norm 3.124, time 6.99s, 
epoch 47, batches 2000|18183, train-acc 0.791, loss 0.516, para-norm 4839.714, grad-norm 2.864, time 6.96s, 
epoch 47, batches 3000|18183, train-acc 0.795, loss 0.515, para-norm 4840.108, grad-norm 5.647, time 6.96s, 
epoch 47, batches 4000|18183, train-acc 0.796, loss 0.503, para-norm 4841.022, grad-norm 4.122, time 6.96s, 
epoch 47, batches 5000|18183, train-acc 0.799, loss 0.508, para-norm 4840.544, grad-norm 4.038, time 6.95s, 
epoch 47, batches 6000|18183, train-acc 0.792, loss 0.513, para-norm 4839.702, grad-norm 2.276, time 6.96s, 
epoch 47, batches 7000|18183, train-acc 0.795, loss 0.515, para-norm 4838.238, grad-norm 8.000, time 6.96s, 
epoch 47, batches 8000|18183, train-acc 0.794, loss 0.514, para-norm 4839.086, grad-norm 6.116, time 6.95s, 
epoch 47, batches 9000|18183, train-acc 0.794, loss 0.518, para-norm 4838.650, grad-norm 5.608, time 6.96s, 
epoch 47, batches 10000|18183, train-acc 0.794, loss 0.512, para-norm 4840.312, grad-norm 11.769, time 6.96s, 
epoch 47, batches 11000|18183, train-acc 0.797, loss 0.506, para-norm 4838.951, grad-norm 2.591, time 6.96s, 
epoch 47, batches 12000|18183, train-acc 0.798, loss 0.506, para-norm 4836.903, grad-norm 2.236, time 6.96s, 
epoch 47, batches 13000|18183, train-acc 0.793, loss 0.514, para-norm 4836.970, grad-norm 7.566, time 6.96s, 
epoch 47, batches 14000|18183, train-acc 0.798, loss 0.506, para-norm 4835.988, grad-norm 2.794, time 6.96s, 
epoch 47, batches 15000|18183, train-acc 0.791, loss 0.521, para-norm 4833.902, grad-norm 6.149, time 6.97s, 
epoch 47, batches 16000|18183, train-acc 0.797, loss 0.507, para-norm 4834.405, grad-norm 2.115, time 6.95s, 
epoch 47, batches 17000|18183, train-acc 0.794, loss 0.510, para-norm 4833.396, grad-norm 4.648, time 6.96s, 
epoch 47, batches 18000|18183, train-acc 0.795, loss 0.512, para-norm 4830.474, grad-norm 5.284, time 6.96s, 
epoch 47, batches 18183|18183, train-acc 0.792, loss 0.511, para-norm 4830.337, grad-norm 1.863, time 1.27s, 
dev-acc 0.821
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
save model!
epoch 48, batches 1000|18183, train-acc 0.793, loss 0.511, para-norm 4830.595, grad-norm 3.707, time 7.21s, 
epoch 48, batches 2000|18183, train-acc 0.792, loss 0.516, para-norm 4831.052, grad-norm 5.472, time 6.96s, 
epoch 48, batches 3000|18183, train-acc 0.801, loss 0.499, para-norm 4831.961, grad-norm 60.621, time 6.95s, 
epoch 48, batches 4000|18183, train-acc 0.793, loss 0.513, para-norm 4831.887, grad-norm 1.627, time 6.96s, 
epoch 48, batches 5000|18183, train-acc 0.800, loss 0.505, para-norm 4832.265, grad-norm 3.006, time 6.95s, 
epoch 48, batches 6000|18183, train-acc 0.797, loss 0.511, para-norm 4831.849, grad-norm 5.849, time 6.95s, 
epoch 48, batches 7000|18183, train-acc 0.799, loss 0.505, para-norm 4831.158, grad-norm 178.897, time 6.94s, 
epoch 48, batches 8000|18183, train-acc 0.798, loss 0.505, para-norm 4830.864, grad-norm 2.098, time 6.96s, 
epoch 48, batches 9000|18183, train-acc 0.797, loss 0.504, para-norm 4830.524, grad-norm 1.292, time 6.96s, 
epoch 48, batches 10000|18183, train-acc 0.792, loss 0.515, para-norm 4830.859, grad-norm 3.163, time 6.95s, 
epoch 48, batches 11000|18183, train-acc 0.795, loss 0.512, para-norm 4831.226, grad-norm 2.593, time 6.96s, 
epoch 48, batches 12000|18183, train-acc 0.796, loss 0.509, para-norm 4833.140, grad-norm 4.381, time 6.96s, 
epoch 48, batches 13000|18183, train-acc 0.796, loss 0.512, para-norm 4831.725, grad-norm 4.521, time 6.96s, 
epoch 48, batches 14000|18183, train-acc 0.796, loss 0.508, para-norm 4830.947, grad-norm 3.647, time 6.96s, 
epoch 48, batches 15000|18183, train-acc 0.796, loss 0.510, para-norm 4830.993, grad-norm 1.609, time 7.16s, 
epoch 48, batches 16000|18183, train-acc 0.794, loss 0.509, para-norm 4830.810, grad-norm 8.531, time 6.96s, 
epoch 48, batches 17000|18183, train-acc 0.799, loss 0.509, para-norm 4829.850, grad-norm 1.729, time 6.95s, 
epoch 48, batches 18000|18183, train-acc 0.798, loss 0.508, para-norm 4828.128, grad-norm 83.244, time 6.96s, 
epoch 48, batches 18183|18183, train-acc 0.796, loss 0.516, para-norm 4828.324, grad-norm 4.634, time 1.27s, 
dev-acc 0.817
epoch 49, batches 1000|18183, train-acc 0.800, loss 0.501, para-norm 4828.853, grad-norm 3.245, time 7.01s, 
epoch 49, batches 2000|18183, train-acc 0.797, loss 0.506, para-norm 4827.912, grad-norm 15.523, time 6.97s, 
epoch 49, batches 3000|18183, train-acc 0.799, loss 0.503, para-norm 4827.591, grad-norm 3.557, time 6.95s, 
epoch 49, batches 4000|18183, train-acc 0.800, loss 0.504, para-norm 4827.495, grad-norm 7.462, time 6.94s, 
epoch 49, batches 5000|18183, train-acc 0.799, loss 0.504, para-norm 4827.506, grad-norm 2.997, time 6.95s, 
epoch 49, batches 6000|18183, train-acc 0.795, loss 0.512, para-norm 4827.466, grad-norm 2.835, time 6.96s, 
epoch 49, batches 7000|18183, train-acc 0.794, loss 0.511, para-norm 4827.872, grad-norm 3.495, time 6.95s, 
epoch 49, batches 8000|18183, train-acc 0.799, loss 0.502, para-norm 4829.493, grad-norm 1.519, time 6.96s, 
epoch 49, batches 9000|18183, train-acc 0.797, loss 0.507, para-norm 4828.697, grad-norm 7.983, time 6.95s, 
epoch 49, batches 10000|18183, train-acc 0.799, loss 0.506, para-norm 4828.797, grad-norm 3.877, time 6.95s, 
epoch 49, batches 11000|18183, train-acc 0.797, loss 0.503, para-norm 4828.749, grad-norm 7.437, time 6.96s, 
epoch 49, batches 12000|18183, train-acc 0.800, loss 0.499, para-norm 4829.437, grad-norm 5.607, time 6.96s, 
epoch 49, batches 13000|18183, train-acc 0.795, loss 0.514, para-norm 4828.877, grad-norm 4.434, time 6.95s, 
epoch 49, batches 14000|18183, train-acc 0.794, loss 0.514, para-norm 4827.860, grad-norm 4.736, time 6.95s, 
epoch 49, batches 15000|18183, train-acc 0.797, loss 0.506, para-norm 4828.649, grad-norm 5.889, time 6.96s, 
epoch 49, batches 16000|18183, train-acc 0.795, loss 0.510, para-norm 4827.294, grad-norm 1.535, time 6.95s, 
epoch 49, batches 17000|18183, train-acc 0.794, loss 0.506, para-norm 4833.379, grad-norm 9.573, time 6.97s, 
epoch 49, batches 18000|18183, train-acc 0.795, loss 0.512, para-norm 4832.943, grad-norm 1.392, time 6.96s, 
epoch 49, batches 18183|18183, train-acc 0.795, loss 0.514, para-norm 4833.076, grad-norm 5.056, time 1.27s, 
dev-acc 0.820
epoch 50, batches 1000|18183, train-acc 0.804, loss 0.499, para-norm 4833.568, grad-norm 2.644, time 7.01s, 
epoch 50, batches 2000|18183, train-acc 0.797, loss 0.504, para-norm 4834.993, grad-norm 4.605, time 6.95s, 
epoch 50, batches 3000|18183, train-acc 0.798, loss 0.505, para-norm 4835.335, grad-norm 2.957, time 6.97s, 
epoch 50, batches 4000|18183, train-acc 0.801, loss 0.503, para-norm 4834.249, grad-norm 7.333, time 6.96s, 
epoch 50, batches 5000|18183, train-acc 0.798, loss 0.505, para-norm 4833.481, grad-norm 6.240, time 6.93s, 
epoch 50, batches 6000|18183, train-acc 0.799, loss 0.505, para-norm 4839.522, grad-norm 1.633, time 6.95s, 
epoch 50, batches 7000|18183, train-acc 0.798, loss 0.506, para-norm 4839.023, grad-norm 6.306, time 6.94s, 
epoch 50, batches 8000|18183, train-acc 0.793, loss 0.514, para-norm 4837.501, grad-norm 8.327, time 6.95s, 
epoch 50, batches 9000|18183, train-acc 0.799, loss 0.505, para-norm 4837.340, grad-norm 1.719, time 6.95s, 
epoch 50, batches 10000|18183, train-acc 0.798, loss 0.499, para-norm 4839.280, grad-norm 6.950, time 6.95s, 
epoch 50, batches 11000|18183, train-acc 0.797, loss 0.505, para-norm 4840.203, grad-norm 4.741, time 6.97s, 
epoch 50, batches 12000|18183, train-acc 0.799, loss 0.501, para-norm 4839.580, grad-norm 13.765, time 6.95s, 
epoch 50, batches 13000|18183, train-acc 0.799, loss 0.505, para-norm 4838.193, grad-norm 4.554, time 6.95s, 
epoch 50, batches 14000|18183, train-acc 0.799, loss 0.498, para-norm 4839.221, grad-norm 3.483, time 6.95s, 
epoch 50, batches 15000|18183, train-acc 0.796, loss 0.507, para-norm 4839.704, grad-norm 5.017, time 6.95s, 
epoch 50, batches 16000|18183, train-acc 0.800, loss 0.500, para-norm 4838.935, grad-norm 2.180, time 6.96s, 
epoch 50, batches 17000|18183, train-acc 0.798, loss 0.506, para-norm 4838.424, grad-norm 6.659, time 6.97s, 
epoch 50, batches 18000|18183, train-acc 0.799, loss 0.500, para-norm 4836.982, grad-norm 2.970, time 6.96s, 
epoch 50, batches 18183|18183, train-acc 0.797, loss 0.506, para-norm 4836.448, grad-norm 6.159, time 1.27s, 
dev-acc 0.821
epoch 51, batches 1000|18183, train-acc 0.802, loss 0.500, para-norm 4837.352, grad-norm 2.517, time 7.09s, 
epoch 51, batches 2000|18183, train-acc 0.804, loss 0.499, para-norm 4837.920, grad-norm 1.641, time 6.94s, 
epoch 51, batches 3000|18183, train-acc 0.797, loss 0.506, para-norm 4837.161, grad-norm 2.691, time 6.97s, 
epoch 51, batches 4000|18183, train-acc 0.801, loss 0.498, para-norm 4837.004, grad-norm 4.815, time 6.95s, 
epoch 51, batches 5000|18183, train-acc 0.797, loss 0.504, para-norm 4837.142, grad-norm 4.172, time 6.97s, 
epoch 51, batches 6000|18183, train-acc 0.801, loss 0.503, para-norm 4835.574, grad-norm 3.001, time 6.96s, 
epoch 51, batches 7000|18183, train-acc 0.799, loss 0.501, para-norm 4841.655, grad-norm 4.247, time 6.96s, 
epoch 51, batches 8000|18183, train-acc 0.798, loss 0.500, para-norm 4841.621, grad-norm 2.251, time 6.97s, 
epoch 51, batches 9000|18183, train-acc 0.798, loss 0.508, para-norm 4842.343, grad-norm 11.977, time 7.02s, 
epoch 51, batches 10000|18183, train-acc 0.796, loss 0.508, para-norm 4842.769, grad-norm 3.674, time 6.98s, 
epoch 51, batches 11000|18183, train-acc 0.804, loss 0.498, para-norm 4843.849, grad-norm 21.857, time 6.96s, 
epoch 51, batches 12000|18183, train-acc 0.798, loss 0.504, para-norm 4844.498, grad-norm 3.636, time 6.96s, 
epoch 51, batches 13000|18183, train-acc 0.802, loss 0.501, para-norm 4843.426, grad-norm 3.015, time 6.95s, 
epoch 51, batches 14000|18183, train-acc 0.802, loss 0.498, para-norm 4843.755, grad-norm 4.853, time 6.96s, 
epoch 51, batches 15000|18183, train-acc 0.798, loss 0.507, para-norm 4852.627, grad-norm 8.727, time 6.97s, 
epoch 51, batches 16000|18183, train-acc 0.801, loss 0.499, para-norm 4850.782, grad-norm 3.743, time 6.95s, 
epoch 51, batches 17000|18183, train-acc 0.798, loss 0.507, para-norm 4851.035, grad-norm 9.388, time 6.97s, 
epoch 51, batches 18000|18183, train-acc 0.798, loss 0.506, para-norm 4855.028, grad-norm 2.630, time 6.96s, 
epoch 51, batches 18183|18183, train-acc 0.797, loss 0.504, para-norm 4855.680, grad-norm 2.832, time 1.27s, 
dev-acc 0.819
epoch 52, batches 1000|18183, train-acc 0.800, loss 0.499, para-norm 4857.135, grad-norm 3.427, time 7.05s, 
epoch 52, batches 2000|18183, train-acc 0.807, loss 0.489, para-norm 4855.365, grad-norm 2.853, time 6.96s, 
epoch 52, batches 3000|18183, train-acc 0.800, loss 0.501, para-norm 4854.153, grad-norm 1.877, time 6.97s, 
epoch 52, batches 4000|18183, train-acc 0.806, loss 0.495, para-norm 4852.955, grad-norm 8.011, time 6.97s, 
epoch 52, batches 5000|18183, train-acc 0.802, loss 0.500, para-norm 4852.207, grad-norm 2.448, time 6.97s, 
epoch 52, batches 6000|18183, train-acc 0.798, loss 0.503, para-norm 4851.946, grad-norm 3.004, time 6.98s, 
epoch 52, batches 7000|18183, train-acc 0.801, loss 0.500, para-norm 4852.437, grad-norm 4.955, time 6.98s, 
epoch 52, batches 8000|18183, train-acc 0.799, loss 0.505, para-norm 4851.091, grad-norm 2.368, time 6.96s, 
epoch 52, batches 9000|18183, train-acc 0.803, loss 0.498, para-norm 4858.952, grad-norm 4.983, time 6.96s, 
epoch 52, batches 10000|18183, train-acc 0.800, loss 0.500, para-norm 4859.302, grad-norm 2.225, time 6.95s, 
epoch 52, batches 11000|18183, train-acc 0.803, loss 0.496, para-norm 4863.979, grad-norm 2079.319, time 6.96s, 
epoch 52, batches 12000|18183, train-acc 0.802, loss 0.496, para-norm 4867.337, grad-norm 2.444, time 6.96s, 
epoch 52, batches 13000|18183, train-acc 0.795, loss 0.511, para-norm 4865.867, grad-norm 5.923, time 6.99s, 
epoch 52, batches 14000|18183, train-acc 0.803, loss 0.501, para-norm 4864.959, grad-norm 2.553, time 6.96s, 
epoch 52, batches 15000|18183, train-acc 0.798, loss 0.506, para-norm 4864.342, grad-norm 4.951, time 6.94s, 
epoch 52, batches 16000|18183, train-acc 0.797, loss 0.506, para-norm 4863.405, grad-norm 2.912, time 6.97s, 
epoch 52, batches 17000|18183, train-acc 0.800, loss 0.501, para-norm 4862.797, grad-norm 2.550, time 6.97s, 
epoch 52, batches 18000|18183, train-acc 0.802, loss 0.500, para-norm 4862.212, grad-norm 3.844, time 7.03s, 
epoch 52, batches 18183|18183, train-acc 0.796, loss 0.505, para-norm 4862.302, grad-norm 2.569, time 1.28s, 
dev-acc 0.822
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
save model!
epoch 53, batches 1000|18183, train-acc 0.801, loss 0.500, para-norm 4862.515, grad-norm 5.240, time 7.00s, 
epoch 53, batches 2000|18183, train-acc 0.804, loss 0.494, para-norm 4862.339, grad-norm 4.736, time 6.96s, 
epoch 53, batches 3000|18183, train-acc 0.802, loss 0.500, para-norm 4861.119, grad-norm 3.544, time 6.95s, 
epoch 53, batches 4000|18183, train-acc 0.803, loss 0.494, para-norm 4860.533, grad-norm 17.893, time 6.96s, 
epoch 53, batches 5000|18183, train-acc 0.801, loss 0.499, para-norm 4861.465, grad-norm 4.411, time 6.97s, 
epoch 53, batches 6000|18183, train-acc 0.802, loss 0.497, para-norm 4863.772, grad-norm 3.581, time 6.95s, 
epoch 53, batches 7000|18183, train-acc 0.801, loss 0.504, para-norm 4862.845, grad-norm 3.305, time 6.96s, 
epoch 53, batches 8000|18183, train-acc 0.802, loss 0.497, para-norm 4861.390, grad-norm 2.582, time 6.97s, 
epoch 53, batches 9000|18183, train-acc 0.800, loss 0.501, para-norm 4860.693, grad-norm 3.812, time 6.95s, 
epoch 53, batches 10000|18183, train-acc 0.802, loss 0.498, para-norm 4859.861, grad-norm 3.976, time 6.95s, 
epoch 53, batches 11000|18183, train-acc 0.805, loss 0.493, para-norm 4859.982, grad-norm 5.109, time 6.96s, 
epoch 53, batches 12000|18183, train-acc 0.803, loss 0.497, para-norm 4859.482, grad-norm 8.200, time 6.95s, 
epoch 53, batches 13000|18183, train-acc 0.801, loss 0.497, para-norm 4868.411, grad-norm 4.270, time 6.96s, 
epoch 53, batches 14000|18183, train-acc 0.800, loss 0.505, para-norm 4868.406, grad-norm 2.807, time 6.95s, 
epoch 53, batches 15000|18183, train-acc 0.798, loss 0.505, para-norm 4868.626, grad-norm 3.442, time 6.95s, 
epoch 53, batches 16000|18183, train-acc 0.800, loss 0.502, para-norm 4871.161, grad-norm 2.708, time 6.95s, 
epoch 53, batches 17000|18183, train-acc 0.803, loss 0.496, para-norm 4870.566, grad-norm 5.329, time 6.95s, 
epoch 53, batches 18000|18183, train-acc 0.799, loss 0.506, para-norm 4868.122, grad-norm 3.435, time 6.96s, 
epoch 53, batches 18183|18183, train-acc 0.795, loss 0.495, para-norm 4867.958, grad-norm 2.463, time 1.28s, 
dev-acc 0.823
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
save model!
epoch 54, batches 1000|18183, train-acc 0.803, loss 0.493, para-norm 4868.383, grad-norm 2.040, time 7.03s, 
epoch 54, batches 2000|18183, train-acc 0.802, loss 0.494, para-norm 4869.589, grad-norm 6.764, time 6.95s, 
epoch 54, batches 3000|18183, train-acc 0.804, loss 0.490, para-norm 4867.587, grad-norm 5.900, time 6.95s, 
epoch 54, batches 4000|18183, train-acc 0.803, loss 0.496, para-norm 4867.671, grad-norm 3.336, time 6.97s, 
epoch 54, batches 5000|18183, train-acc 0.802, loss 0.494, para-norm 4868.013, grad-norm 7.668, time 6.95s, 
epoch 54, batches 6000|18183, train-acc 0.800, loss 0.502, para-norm 4865.132, grad-norm 2.832, time 6.97s, 
epoch 54, batches 7000|18183, train-acc 0.800, loss 0.500, para-norm 4866.237, grad-norm 5.559, time 6.97s, 
epoch 54, batches 8000|18183, train-acc 0.801, loss 0.494, para-norm 4865.862, grad-norm 33.471, time 6.96s, 
epoch 54, batches 9000|18183, train-acc 0.802, loss 0.496, para-norm 4864.439, grad-norm 5.545, time 6.95s, 
epoch 54, batches 10000|18183, train-acc 0.796, loss 0.508, para-norm 4863.306, grad-norm 2.677, time 6.97s, 
epoch 54, batches 11000|18183, train-acc 0.799, loss 0.501, para-norm 4863.324, grad-norm 3.230, time 6.96s, 
epoch 54, batches 12000|18183, train-acc 0.798, loss 0.507, para-norm 4862.384, grad-norm 1.848, time 6.97s, 
epoch 54, batches 13000|18183, train-acc 0.800, loss 0.500, para-norm 4861.916, grad-norm 5.058, time 6.97s, 
epoch 54, batches 14000|18183, train-acc 0.805, loss 0.491, para-norm 4862.360, grad-norm 5.750, time 6.95s, 
epoch 54, batches 15000|18183, train-acc 0.805, loss 0.490, para-norm 4863.216, grad-norm 6.912, time 6.96s, 
epoch 54, batches 16000|18183, train-acc 0.804, loss 0.495, para-norm 4861.918, grad-norm 4.543, time 6.97s, 
epoch 54, batches 17000|18183, train-acc 0.801, loss 0.503, para-norm 4861.723, grad-norm 5.112, time 6.95s, 
epoch 54, batches 18000|18183, train-acc 0.807, loss 0.489, para-norm 4862.052, grad-norm 2.494, time 6.95s, 
epoch 54, batches 18183|18183, train-acc 0.802, loss 0.503, para-norm 4862.089, grad-norm 2.756, time 1.27s, 
dev-acc 0.826
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
save model!
epoch 55, batches 1000|18183, train-acc 0.810, loss 0.482, para-norm 4864.766, grad-norm 2.313, time 7.04s, 
epoch 55, batches 2000|18183, train-acc 0.808, loss 0.487, para-norm 4863.957, grad-norm 3.793, time 6.96s, 
epoch 55, batches 3000|18183, train-acc 0.809, loss 0.488, para-norm 4863.454, grad-norm 12.030, time 6.96s, 
epoch 55, batches 4000|18183, train-acc 0.802, loss 0.493, para-norm 4863.320, grad-norm 4.792, time 6.96s, 
epoch 55, batches 5000|18183, train-acc 0.801, loss 0.498, para-norm 4862.891, grad-norm 0.004, time 6.95s, 
epoch 55, batches 6000|18183, train-acc 0.799, loss 0.501, para-norm 4861.171, grad-norm 5.101, time 6.96s, 
epoch 55, batches 7000|18183, train-acc 0.805, loss 0.489, para-norm 4861.107, grad-norm 3.420, time 6.97s, 
epoch 55, batches 8000|18183, train-acc 0.799, loss 0.498, para-norm 4861.092, grad-norm 10.703, time 6.96s, 
epoch 55, batches 9000|18183, train-acc 0.802, loss 0.499, para-norm 4861.054, grad-norm 2.491, time 6.94s, 
epoch 55, batches 10000|18183, train-acc 0.805, loss 0.491, para-norm 4866.724, grad-norm 4.281, time 7.31s, 
epoch 55, batches 11000|18183, train-acc 0.804, loss 0.492, para-norm 4871.361, grad-norm 1.511, time 7.76s, 
epoch 55, batches 12000|18183, train-acc 0.801, loss 0.496, para-norm 4870.037, grad-norm 2.332, time 7.78s, 
epoch 55, batches 13000|18183, train-acc 0.801, loss 0.501, para-norm 4868.641, grad-norm 3.815, time 7.78s, 
epoch 55, batches 14000|18183, train-acc 0.810, loss 0.488, para-norm 4868.974, grad-norm 3.711, time 7.76s, 
epoch 55, batches 15000|18183, train-acc 0.802, loss 0.500, para-norm 4867.163, grad-norm 4.950, time 7.80s, 
epoch 55, batches 16000|18183, train-acc 0.799, loss 0.502, para-norm 4866.139, grad-norm 5.334, time 7.78s, 
epoch 55, batches 17000|18183, train-acc 0.802, loss 0.495, para-norm 4865.047, grad-norm 2.458, time 7.76s, 
epoch 55, batches 18000|18183, train-acc 0.804, loss 0.494, para-norm 4864.096, grad-norm 4.969, time 7.78s, 
epoch 55, batches 18183|18183, train-acc 0.802, loss 0.510, para-norm 4863.953, grad-norm 6.801, time 1.43s, 
dev-acc 0.825
epoch 56, batches 1000|18183, train-acc 0.806, loss 0.488, para-norm 4863.950, grad-norm 3.842, time 7.91s, 
epoch 56, batches 2000|18183, train-acc 0.802, loss 0.493, para-norm 4863.875, grad-norm 4.500, time 7.78s, 
epoch 56, batches 3000|18183, train-acc 0.805, loss 0.492, para-norm 4863.326, grad-norm 2.834, time 7.77s, 
epoch 56, batches 4000|18183, train-acc 0.803, loss 0.491, para-norm 4863.206, grad-norm 6.541, time 7.77s, 
epoch 56, batches 5000|18183, train-acc 0.808, loss 0.488, para-norm 4863.389, grad-norm 3.355, time 7.78s, 
epoch 56, batches 6000|18183, train-acc 0.802, loss 0.493, para-norm 4871.712, grad-norm 3.538, time 7.78s, 
epoch 56, batches 7000|18183, train-acc 0.805, loss 0.493, para-norm 4870.575, grad-norm 2.362, time 7.78s, 
epoch 56, batches 8000|18183, train-acc 0.803, loss 0.494, para-norm 4870.611, grad-norm 1.350, time 7.78s, 
epoch 56, batches 9000|18183, train-acc 0.808, loss 0.491, para-norm 4869.766, grad-norm 2.970, time 7.78s, 
epoch 56, batches 10000|18183, train-acc 0.805, loss 0.491, para-norm 4877.274, grad-norm 1.550, time 7.77s, 
epoch 56, batches 11000|18183, train-acc 0.807, loss 0.492, para-norm 4876.931, grad-norm 3.373, time 7.78s, 
epoch 56, batches 12000|18183, train-acc 0.803, loss 0.492, para-norm 4876.230, grad-norm 1.581, time 7.78s, 
epoch 56, batches 13000|18183, train-acc 0.803, loss 0.498, para-norm 4875.692, grad-norm 4.281, time 7.79s, 
epoch 56, batches 14000|18183, train-acc 0.804, loss 0.493, para-norm 4874.391, grad-norm 5.144, time 7.78s, 
epoch 56, batches 15000|18183, train-acc 0.804, loss 0.490, para-norm 4874.061, grad-norm 2.454, time 7.77s, 
epoch 56, batches 16000|18183, train-acc 0.803, loss 0.496, para-norm 4873.533, grad-norm 2.524, time 7.77s, 
epoch 56, batches 17000|18183, train-acc 0.805, loss 0.494, para-norm 4872.675, grad-norm 3.126, time 7.76s, 
epoch 56, batches 18000|18183, train-acc 0.803, loss 0.492, para-norm 4873.180, grad-norm 2.012, time 7.77s, 
epoch 56, batches 18183|18183, train-acc 0.798, loss 0.504, para-norm 4872.955, grad-norm 2.064, time 1.42s, 
dev-acc 0.823
epoch 57, batches 1000|18183, train-acc 0.809, loss 0.487, para-norm 4873.154, grad-norm 5.274, time 7.86s, 
epoch 57, batches 2000|18183, train-acc 0.803, loss 0.497, para-norm 4873.113, grad-norm 2.399, time 7.80s, 
epoch 57, batches 3000|18183, train-acc 0.805, loss 0.488, para-norm 4872.425, grad-norm 2.380, time 7.79s, 
epoch 57, batches 4000|18183, train-acc 0.807, loss 0.486, para-norm 4872.493, grad-norm 4.869, time 7.79s, 
epoch 57, batches 5000|18183, train-acc 0.809, loss 0.484, para-norm 4872.357, grad-norm 3.977, time 7.79s, 
epoch 57, batches 6000|18183, train-acc 0.807, loss 0.485, para-norm 4873.172, grad-norm 3.544, time 7.77s, 
epoch 57, batches 7000|18183, train-acc 0.806, loss 0.486, para-norm 4872.349, grad-norm 0.000, time 7.77s, 
epoch 57, batches 8000|18183, train-acc 0.810, loss 0.477, para-norm 4872.650, grad-norm 2.373, time 7.78s, 
epoch 57, batches 9000|18183, train-acc 0.802, loss 0.493, para-norm 4873.474, grad-norm 1.952, time 7.77s, 
epoch 57, batches 10000|18183, train-acc 0.806, loss 0.491, para-norm 4873.402, grad-norm 2.588, time 7.77s, 
epoch 57, batches 11000|18183, train-acc 0.808, loss 0.483, para-norm 4872.490, grad-norm 3.760, time 7.77s, 
epoch 57, batches 12000|18183, train-acc 0.805, loss 0.485, para-norm 4873.362, grad-norm 3.784, time 7.75s, 
epoch 57, batches 13000|18183, train-acc 0.807, loss 0.485, para-norm 4872.407, grad-norm 5.155, time 7.78s, 
epoch 57, batches 14000|18183, train-acc 0.805, loss 0.491, para-norm 4870.951, grad-norm 3.739, time 7.78s, 
epoch 57, batches 15000|18183, train-acc 0.804, loss 0.497, para-norm 4869.876, grad-norm 3.112, time 7.76s, 
epoch 57, batches 16000|18183, train-acc 0.804, loss 0.490, para-norm 4868.402, grad-norm 2.341, time 7.78s, 
epoch 57, batches 17000|18183, train-acc 0.804, loss 0.492, para-norm 4869.143, grad-norm 3.981, time 7.77s, 
epoch 57, batches 18000|18183, train-acc 0.804, loss 0.492, para-norm 4868.808, grad-norm 3.340, time 7.77s, 
epoch 57, batches 18183|18183, train-acc 0.801, loss 0.500, para-norm 4868.725, grad-norm 7.771, time 1.43s, 
dev-acc 0.824
epoch 58, batches 1000|18183, train-acc 0.808, loss 0.483, para-norm 4869.958, grad-norm 4.692, time 7.84s, 
epoch 58, batches 2000|18183, train-acc 0.805, loss 0.488, para-norm 4869.229, grad-norm 4.847, time 7.79s, 
epoch 58, batches 3000|18183, train-acc 0.801, loss 0.495, para-norm 4867.547, grad-norm 6.487, time 7.78s, 
epoch 58, batches 4000|18183, train-acc 0.812, loss 0.480, para-norm 4868.747, grad-norm 4.113, time 7.75s, 
epoch 58, batches 5000|18183, train-acc 0.806, loss 0.492, para-norm 4867.955, grad-norm 3.867, time 7.78s, 
epoch 58, batches 6000|18183, train-acc 0.807, loss 0.484, para-norm 4867.786, grad-norm 5.110, time 7.77s, 
epoch 58, batches 7000|18183, train-acc 0.802, loss 0.493, para-norm 4867.332, grad-norm 6.027, time 7.77s, 
epoch 58, batches 8000|18183, train-acc 0.809, loss 0.487, para-norm 4866.128, grad-norm 2.535, time 7.78s, 
epoch 58, batches 9000|18183, train-acc 0.806, loss 0.490, para-norm 4865.595, grad-norm 1.910, time 7.76s, 
epoch 58, batches 10000|18183, train-acc 0.807, loss 0.487, para-norm 4865.219, grad-norm 4.028, time 7.78s, 
epoch 58, batches 11000|18183, train-acc 0.808, loss 0.485, para-norm 4865.148, grad-norm 3.637, time 7.78s, 
epoch 58, batches 12000|18183, train-acc 0.805, loss 0.489, para-norm 4864.887, grad-norm 14.151, time 7.76s, 
epoch 58, batches 13000|18183, train-acc 0.806, loss 0.490, para-norm 4864.016, grad-norm 1.849, time 7.77s, 
epoch 58, batches 14000|18183, train-acc 0.808, loss 0.483, para-norm 4863.509, grad-norm 2.778, time 7.79s, 
epoch 58, batches 15000|18183, train-acc 0.804, loss 0.490, para-norm 4863.347, grad-norm 3.324, time 7.77s, 
epoch 58, batches 16000|18183, train-acc 0.810, loss 0.479, para-norm 4862.892, grad-norm 2.414, time 7.75s, 
epoch 58, batches 17000|18183, train-acc 0.808, loss 0.485, para-norm 4862.097, grad-norm 4.072, time 7.77s, 
epoch 58, batches 18000|18183, train-acc 0.806, loss 0.488, para-norm 4862.509, grad-norm 2.604, time 7.77s, 
epoch 58, batches 18183|18183, train-acc 0.802, loss 0.498, para-norm 4862.392, grad-norm 2.440, time 1.42s, 
dev-acc 0.829
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
save model!
epoch 59, batches 1000|18183, train-acc 0.805, loss 0.495, para-norm 4862.349, grad-norm 5.212, time 7.31s, 
epoch 59, batches 2000|18183, train-acc 0.812, loss 0.474, para-norm 4862.122, grad-norm 2.839, time 6.95s, 
epoch 59, batches 3000|18183, train-acc 0.811, loss 0.480, para-norm 4861.863, grad-norm 4.309, time 6.95s, 
epoch 59, batches 4000|18183, train-acc 0.806, loss 0.483, para-norm 4862.038, grad-norm 2.813, time 6.96s, 
epoch 59, batches 5000|18183, train-acc 0.807, loss 0.491, para-norm 4861.473, grad-norm 3.568, time 6.95s, 
epoch 59, batches 6000|18183, train-acc 0.810, loss 0.481, para-norm 4860.561, grad-norm 3.581, time 6.96s, 
epoch 59, batches 7000|18183, train-acc 0.811, loss 0.477, para-norm 4860.420, grad-norm 6.981, time 6.94s, 
epoch 59, batches 8000|18183, train-acc 0.805, loss 0.491, para-norm 4860.426, grad-norm 3.128, time 6.94s, 
epoch 59, batches 9000|18183, train-acc 0.809, loss 0.482, para-norm 4859.799, grad-norm 5.609, time 6.96s, 
epoch 59, batches 10000|18183, train-acc 0.805, loss 0.491, para-norm 4858.839, grad-norm 3.279, time 6.97s, 
epoch 59, batches 11000|18183, train-acc 0.810, loss 0.484, para-norm 4858.076, grad-norm 5.210, time 6.95s, 
epoch 59, batches 12000|18183, train-acc 0.807, loss 0.481, para-norm 4857.113, grad-norm 3.085, time 6.95s, 
epoch 59, batches 13000|18183, train-acc 0.809, loss 0.483, para-norm 4856.468, grad-norm 3.794, time 6.95s, 
epoch 59, batches 14000|18183, train-acc 0.809, loss 0.480, para-norm 4855.762, grad-norm 4.289, time 6.96s, 
epoch 59, batches 15000|18183, train-acc 0.803, loss 0.489, para-norm 4855.895, grad-norm 3.743, time 6.95s, 
epoch 59, batches 16000|18183, train-acc 0.808, loss 0.486, para-norm 4854.507, grad-norm 4.863, time 6.96s, 
epoch 59, batches 17000|18183, train-acc 0.806, loss 0.481, para-norm 4854.200, grad-norm 6.047, time 6.96s, 
epoch 59, batches 18000|18183, train-acc 0.810, loss 0.484, para-norm 4853.585, grad-norm 2.974, time 6.94s, 
epoch 59, batches 18183|18183, train-acc 0.802, loss 0.497, para-norm 4853.615, grad-norm 2.979, time 1.28s, 
dev-acc 0.826
epoch 60, batches 1000|18183, train-acc 0.811, loss 0.479, para-norm 4854.134, grad-norm 2.322, time 7.00s, 
epoch 60, batches 2000|18183, train-acc 0.811, loss 0.480, para-norm 4853.264, grad-norm 4.277, time 6.95s, 
epoch 60, batches 3000|18183, train-acc 0.811, loss 0.476, para-norm 4853.398, grad-norm 2.984, time 6.95s, 
epoch 60, batches 4000|18183, train-acc 0.812, loss 0.478, para-norm 4851.953, grad-norm 2.141, time 7.20s, 
epoch 60, batches 5000|18183, train-acc 0.811, loss 0.478, para-norm 4852.181, grad-norm 4.011, time 7.29s, 
epoch 60, batches 6000|18183, train-acc 0.808, loss 0.477, para-norm 4852.051, grad-norm 2.278, time 7.20s, 
epoch 60, batches 7000|18183, train-acc 0.812, loss 0.481, para-norm 4851.104, grad-norm 1.781, time 7.20s, 
epoch 60, batches 8000|18183, train-acc 0.808, loss 0.487, para-norm 4851.204, grad-norm 1.659, time 7.27s, 
epoch 60, batches 9000|18183, train-acc 0.811, loss 0.477, para-norm 4858.113, grad-norm 4.648, time 7.21s, 
epoch 60, batches 10000|18183, train-acc 0.808, loss 0.485, para-norm 4858.255, grad-norm 2.680, time 7.20s, 
epoch 60, batches 11000|18183, train-acc 0.808, loss 0.483, para-norm 4859.089, grad-norm 1.464, time 7.20s, 
epoch 60, batches 12000|18183, train-acc 0.806, loss 0.485, para-norm 4857.748, grad-norm 3.573, time 7.20s, 
epoch 60, batches 13000|18183, train-acc 0.807, loss 0.484, para-norm 4857.340, grad-norm 6.749, time 7.19s, 
epoch 60, batches 14000|18183, train-acc 0.811, loss 0.481, para-norm 4856.132, grad-norm 3.590, time 7.21s, 
epoch 60, batches 15000|18183, train-acc 0.810, loss 0.480, para-norm 4855.627, grad-norm 4.509, time 7.20s, 
epoch 60, batches 16000|18183, train-acc 0.807, loss 0.485, para-norm 4854.665, grad-norm 2.954, time 7.19s, 
epoch 60, batches 17000|18183, train-acc 0.808, loss 0.484, para-norm 4860.269, grad-norm 4.968, time 7.21s, 
epoch 60, batches 18000|18183, train-acc 0.807, loss 0.482, para-norm 4860.267, grad-norm 287.684, time 7.21s, 
epoch 60, batches 18183|18183, train-acc 0.804, loss 0.489, para-norm 4859.998, grad-norm 5.170, time 1.32s, 
dev-acc 0.831
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
save model!
epoch 61, batches 1000|18183, train-acc 0.809, loss 0.486, para-norm 4866.211, grad-norm 2.761, time 7.27s, 
epoch 61, batches 2000|18183, train-acc 0.812, loss 0.477, para-norm 4871.858, grad-norm 4.024, time 7.20s, 
epoch 61, batches 3000|18183, train-acc 0.812, loss 0.476, para-norm 4875.028, grad-norm 5.498, time 7.19s, 
epoch 61, batches 4000|18183, train-acc 0.810, loss 0.479, para-norm 4872.993, grad-norm 2.953, time 7.22s, 
epoch 61, batches 5000|18183, train-acc 0.809, loss 0.480, para-norm 4872.591, grad-norm 2.590, time 7.20s, 
epoch 61, batches 6000|18183, train-acc 0.809, loss 0.483, para-norm 4871.762, grad-norm 2.927, time 7.21s, 
epoch 61, batches 7000|18183, train-acc 0.811, loss 0.479, para-norm 4871.182, grad-norm 2.709, time 7.23s, 
epoch 61, batches 8000|18183, train-acc 0.810, loss 0.476, para-norm 4870.427, grad-norm 6.948, time 7.20s, 
epoch 61, batches 9000|18183, train-acc 0.807, loss 0.482, para-norm 4868.813, grad-norm 3.541, time 7.21s, 
epoch 61, batches 10000|18183, train-acc 0.810, loss 0.481, para-norm 4867.349, grad-norm 2.766, time 7.20s, 
epoch 61, batches 11000|18183, train-acc 0.814, loss 0.476, para-norm 4866.989, grad-norm 2.755, time 7.21s, 
epoch 61, batches 12000|18183, train-acc 0.811, loss 0.478, para-norm 4874.136, grad-norm 1.650, time 7.22s, 
epoch 61, batches 13000|18183, train-acc 0.813, loss 0.478, para-norm 4872.614, grad-norm 6.298, time 7.20s, 
epoch 61, batches 14000|18183, train-acc 0.811, loss 0.480, para-norm 4870.320, grad-norm 2.978, time 7.21s, 
epoch 61, batches 15000|18183, train-acc 0.814, loss 0.471, para-norm 4870.643, grad-norm 4.383, time 7.21s, 
epoch 61, batches 16000|18183, train-acc 0.808, loss 0.482, para-norm 4869.061, grad-norm 2.205, time 7.21s, 
epoch 61, batches 17000|18183, train-acc 0.812, loss 0.475, para-norm 4868.037, grad-norm 4.554, time 7.20s, 
epoch 61, batches 18000|18183, train-acc 0.807, loss 0.486, para-norm 4870.664, grad-norm 1.202, time 7.21s, 
epoch 61, batches 18183|18183, train-acc 0.814, loss 0.469, para-norm 4870.912, grad-norm 3.805, time 1.31s, 
dev-acc 0.833
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
save model!
epoch 62, batches 1000|18183, train-acc 0.816, loss 0.469, para-norm 4880.683, grad-norm 2.631, time 7.25s, 
epoch 62, batches 2000|18183, train-acc 0.814, loss 0.464, para-norm 4883.526, grad-norm 2.596, time 7.22s, 
epoch 62, batches 3000|18183, train-acc 0.818, loss 0.466, para-norm 4881.610, grad-norm 408.352, time 7.20s, 
epoch 62, batches 4000|18183, train-acc 0.809, loss 0.482, para-norm 4881.890, grad-norm 51.515, time 7.24s, 
epoch 62, batches 5000|18183, train-acc 0.815, loss 0.467, para-norm 4881.612, grad-norm 6.751, time 7.20s, 
epoch 62, batches 6000|18183, train-acc 0.807, loss 0.484, para-norm 4881.194, grad-norm 3.447, time 7.20s, 
epoch 62, batches 7000|18183, train-acc 0.801, loss 0.494, para-norm 4880.540, grad-norm 2.259, time 7.24s, 
epoch 62, batches 8000|18183, train-acc 0.813, loss 0.477, para-norm 4880.252, grad-norm 2.603, time 7.20s, 
epoch 62, batches 9000|18183, train-acc 0.808, loss 0.483, para-norm 4879.850, grad-norm 3.819, time 7.20s, 
epoch 62, batches 10000|18183, train-acc 0.810, loss 0.478, para-norm 4879.086, grad-norm 4.783, time 7.23s, 
epoch 62, batches 11000|18183, train-acc 0.809, loss 0.476, para-norm 4877.794, grad-norm 4.420, time 7.19s, 
epoch 62, batches 12000|18183, train-acc 0.810, loss 0.478, para-norm 4876.824, grad-norm 5.735, time 7.25s, 
epoch 62, batches 13000|18183, train-acc 0.812, loss 0.474, para-norm 4878.111, grad-norm 2.586, time 7.21s, 
epoch 62, batches 14000|18183, train-acc 0.813, loss 0.475, para-norm 4877.646, grad-norm 3.132, time 7.20s, 
epoch 62, batches 15000|18183, train-acc 0.814, loss 0.472, para-norm 4876.562, grad-norm 2.585, time 7.24s, 
epoch 62, batches 16000|18183, train-acc 0.814, loss 0.476, para-norm 4876.190, grad-norm 3.677, time 7.20s, 
epoch 62, batches 17000|18183, train-acc 0.810, loss 0.478, para-norm 4875.399, grad-norm 4.190, time 7.20s, 
epoch 62, batches 18000|18183, train-acc 0.809, loss 0.481, para-norm 4874.845, grad-norm 2.401, time 7.24s, 
epoch 62, batches 18183|18183, train-acc 0.813, loss 0.477, para-norm 4874.698, grad-norm 1.549, time 1.31s, 
dev-acc 0.831
epoch 63, batches 1000|18183, train-acc 0.814, loss 0.471, para-norm 4876.274, grad-norm 3.371, time 7.25s, 
epoch 63, batches 2000|18183, train-acc 0.817, loss 0.469, para-norm 4876.102, grad-norm 4.790, time 7.24s, 
epoch 63, batches 3000|18183, train-acc 0.811, loss 0.474, para-norm 4875.318, grad-norm 4.603, time 7.20s, 
epoch 63, batches 4000|18183, train-acc 0.810, loss 0.482, para-norm 4874.087, grad-norm 1.325, time 7.20s, 
epoch 63, batches 5000|18183, train-acc 0.814, loss 0.474, para-norm 4874.153, grad-norm 3.145, time 7.23s, 
epoch 63, batches 6000|18183, train-acc 0.816, loss 0.468, para-norm 4873.753, grad-norm 3.320, time 7.21s, 
epoch 63, batches 7000|18183, train-acc 0.816, loss 0.469, para-norm 4879.489, grad-norm 2.288, time 7.32s, 
epoch 63, batches 8000|18183, train-acc 0.814, loss 0.470, para-norm 4880.062, grad-norm 1.338, time 7.21s, 
epoch 63, batches 9000|18183, train-acc 0.811, loss 0.477, para-norm 4879.022, grad-norm 4.336, time 7.19s, 
epoch 63, batches 10000|18183, train-acc 0.813, loss 0.472, para-norm 4878.386, grad-norm 2.304, time 7.24s, 
epoch 63, batches 11000|18183, train-acc 0.813, loss 0.468, para-norm 4878.506, grad-norm 3.565, time 7.19s, 
epoch 63, batches 12000|18183, train-acc 0.812, loss 0.475, para-norm 4877.970, grad-norm 2.140, time 7.20s, 
epoch 63, batches 13000|18183, train-acc 0.810, loss 0.480, para-norm 4877.264, grad-norm 2.119, time 7.24s, 
epoch 63, batches 14000|18183, train-acc 0.812, loss 0.475, para-norm 4875.336, grad-norm 4.827, time 7.20s, 
epoch 63, batches 15000|18183, train-acc 0.810, loss 0.479, para-norm 4875.071, grad-norm 488.297, time 7.21s, 
epoch 63, batches 16000|18183, train-acc 0.811, loss 0.477, para-norm 4873.635, grad-norm 1.928, time 7.20s, 
epoch 63, batches 17000|18183, train-acc 0.813, loss 0.472, para-norm 4872.446, grad-norm 2.008, time 7.19s, 
epoch 63, batches 18000|18183, train-acc 0.814, loss 0.471, para-norm 4871.243, grad-norm 1.537, time 7.23s, 
epoch 63, batches 18183|18183, train-acc 0.816, loss 0.470, para-norm 4871.037, grad-norm 3.251, time 1.31s, 
dev-acc 0.833
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
save model!
epoch 64, batches 1000|18183, train-acc 0.818, loss 0.461, para-norm 4870.766, grad-norm 1.915, time 7.33s, 
epoch 64, batches 2000|18183, train-acc 0.816, loss 0.466, para-norm 4873.753, grad-norm 4.003, time 7.19s, 
epoch 64, batches 3000|18183, train-acc 0.813, loss 0.474, para-norm 4872.329, grad-norm 4.981, time 7.19s, 
epoch 64, batches 4000|18183, train-acc 0.816, loss 0.470, para-norm 4874.244, grad-norm 3.874, time 7.28s, 
epoch 64, batches 5000|18183, train-acc 0.813, loss 0.469, para-norm 4873.348, grad-norm 4.410, time 7.27s, 
epoch 64, batches 6000|18183, train-acc 0.814, loss 0.465, para-norm 4874.422, grad-norm 2.414, time 7.19s, 
epoch 64, batches 7000|18183, train-acc 0.819, loss 0.462, para-norm 4875.789, grad-norm 7.201, time 7.18s, 
epoch 64, batches 8000|18183, train-acc 0.813, loss 0.472, para-norm 4874.063, grad-norm 3.293, time 7.19s, 
epoch 64, batches 9000|18183, train-acc 0.816, loss 0.467, para-norm 4873.641, grad-norm 4.620, time 7.18s, 
epoch 64, batches 10000|18183, train-acc 0.816, loss 0.466, para-norm 4872.933, grad-norm 18.866, time 7.20s, 
epoch 64, batches 11000|18183, train-acc 0.815, loss 0.473, para-norm 4871.611, grad-norm 2.777, time 7.20s, 
epoch 64, batches 12000|18183, train-acc 0.811, loss 0.475, para-norm 4870.534, grad-norm 3.011, time 7.20s, 
epoch 64, batches 13000|18183, train-acc 0.815, loss 0.469, para-norm 4870.070, grad-norm 3.411, time 7.18s, 
epoch 64, batches 14000|18183, train-acc 0.813, loss 0.471, para-norm 4870.235, grad-norm 3.647, time 7.21s, 
epoch 64, batches 15000|18183, train-acc 0.817, loss 0.465, para-norm 4868.469, grad-norm 2.410, time 7.19s, 
epoch 64, batches 16000|18183, train-acc 0.814, loss 0.475, para-norm 4867.384, grad-norm 4.444, time 7.17s, 
epoch 64, batches 17000|18183, train-acc 0.817, loss 0.468, para-norm 4866.835, grad-norm 1.292, time 7.19s, 
epoch 64, batches 18000|18183, train-acc 0.813, loss 0.472, para-norm 4866.344, grad-norm 1.753, time 7.20s, 
epoch 64, batches 18183|18183, train-acc 0.815, loss 0.471, para-norm 4867.007, grad-norm 3.023, time 1.31s, 
dev-acc 0.837
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
save model!
epoch 65, batches 1000|18183, train-acc 0.810, loss 0.474, para-norm 4873.700, grad-norm 4.275, time 7.25s, 
epoch 65, batches 2000|18183, train-acc 0.817, loss 0.466, para-norm 4874.153, grad-norm 3.209, time 7.19s, 
epoch 65, batches 3000|18183, train-acc 0.819, loss 0.460, para-norm 4875.095, grad-norm 3.630, time 7.19s, 
epoch 65, batches 4000|18183, train-acc 0.818, loss 0.463, para-norm 4883.509, grad-norm 2.850, time 7.18s, 
epoch 65, batches 5000|18183, train-acc 0.816, loss 0.467, para-norm 4881.495, grad-norm 1.918, time 7.17s, 
epoch 65, batches 6000|18183, train-acc 0.818, loss 0.463, para-norm 4880.597, grad-norm 5.929, time 7.19s, 
epoch 65, batches 7000|18183, train-acc 0.819, loss 0.467, para-norm 4880.595, grad-norm 2.628, time 7.18s, 
epoch 65, batches 8000|18183, train-acc 0.816, loss 0.465, para-norm 4879.533, grad-norm 2.507, time 7.19s, 
epoch 65, batches 9000|18183, train-acc 0.815, loss 0.466, para-norm 4879.567, grad-norm 4.095, time 7.20s, 
epoch 65, batches 10000|18183, train-acc 0.816, loss 0.465, para-norm 4877.540, grad-norm 3.410, time 7.18s, 
epoch 65, batches 11000|18183, train-acc 0.817, loss 0.463, para-norm 4877.160, grad-norm 2.472, time 7.18s, 
epoch 65, batches 12000|18183, train-acc 0.816, loss 0.463, para-norm 4876.142, grad-norm 1.927, time 7.19s, 
epoch 65, batches 13000|18183, train-acc 0.815, loss 0.466, para-norm 4876.140, grad-norm 4.586, time 7.20s, 
epoch 65, batches 14000|18183, train-acc 0.815, loss 0.467, para-norm 4875.611, grad-norm 2.585, time 7.18s, 
epoch 65, batches 15000|18183, train-acc 0.819, loss 0.461, para-norm 4875.816, grad-norm 2.686, time 7.19s, 
epoch 65, batches 16000|18183, train-acc 0.813, loss 0.474, para-norm 4875.426, grad-norm 1.722, time 7.18s, 
epoch 65, batches 17000|18183, train-acc 0.814, loss 0.472, para-norm 4874.603, grad-norm 5.507, time 7.18s, 
epoch 65, batches 18000|18183, train-acc 0.817, loss 0.463, para-norm 4874.466, grad-norm 9.094, time 7.19s, 
epoch 65, batches 18183|18183, train-acc 0.820, loss 0.451, para-norm 4873.396, grad-norm 1.061, time 1.31s, 
dev-acc 0.837
epoch 66, batches 1000|18183, train-acc 0.814, loss 0.468, para-norm 4881.662, grad-norm 2.604, time 7.25s, 
epoch 66, batches 2000|18183, train-acc 0.815, loss 0.467, para-norm 4881.100, grad-norm 3.239, time 7.24s, 
epoch 66, batches 3000|18183, train-acc 0.819, loss 0.458, para-norm 4881.284, grad-norm 2.910, time 7.19s, 
epoch 66, batches 4000|18183, train-acc 0.817, loss 0.460, para-norm 4881.408, grad-norm 1.385, time 7.19s, 
epoch 66, batches 5000|18183, train-acc 0.816, loss 0.461, para-norm 4880.529, grad-norm 2.277, time 7.20s, 
epoch 66, batches 6000|18183, train-acc 0.816, loss 0.467, para-norm 4879.529, grad-norm 3.416, time 7.20s, 
epoch 66, batches 7000|18183, train-acc 0.821, loss 0.453, para-norm 4879.321, grad-norm 2.532, time 7.20s, 
epoch 66, batches 8000|18183, train-acc 0.820, loss 0.458, para-norm 4878.275, grad-norm 1.354, time 7.27s, 
epoch 66, batches 9000|18183, train-acc 0.822, loss 0.456, para-norm 4877.287, grad-norm 1.928, time 7.20s, 
epoch 66, batches 10000|18183, train-acc 0.820, loss 0.458, para-norm 4876.742, grad-norm 4.026, time 7.18s, 
epoch 66, batches 11000|18183, train-acc 0.819, loss 0.467, para-norm 4882.733, grad-norm 1.729, time 7.20s, 
epoch 66, batches 12000|18183, train-acc 0.821, loss 0.464, para-norm 4882.526, grad-norm 2.389, time 7.19s, 
epoch 66, batches 13000|18183, train-acc 0.817, loss 0.465, para-norm 4882.280, grad-norm 3.358, time 7.19s, 
epoch 66, batches 14000|18183, train-acc 0.818, loss 0.467, para-norm 4882.446, grad-norm 4.297, time 7.19s, 
epoch 66, batches 15000|18183, train-acc 0.819, loss 0.464, para-norm 4881.424, grad-norm 3.011, time 7.20s, 
epoch 66, batches 16000|18183, train-acc 0.817, loss 0.458, para-norm 4880.696, grad-norm 7.025, time 7.19s, 
epoch 66, batches 17000|18183, train-acc 0.815, loss 0.471, para-norm 4880.199, grad-norm 21.700, time 7.18s, 
epoch 66, batches 18000|18183, train-acc 0.816, loss 0.466, para-norm 4878.565, grad-norm 2.864, time 7.20s, 
epoch 66, batches 18183|18183, train-acc 0.816, loss 0.465, para-norm 4878.830, grad-norm 3.553, time 1.32s, 
dev-acc 0.842
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
save model!
epoch 67, batches 1000|18183, train-acc 0.819, loss 0.461, para-norm 4878.641, grad-norm 1.797, time 7.40s, 
epoch 67, batches 2000|18183, train-acc 0.821, loss 0.456, para-norm 4877.850, grad-norm 2.024, time 7.18s, 
epoch 67, batches 3000|18183, train-acc 0.818, loss 0.459, para-norm 4877.410, grad-norm 4.020, time 7.20s, 
epoch 67, batches 4000|18183, train-acc 0.816, loss 0.460, para-norm 4876.875, grad-norm 3.518, time 7.18s, 
epoch 67, batches 5000|18183, train-acc 0.813, loss 0.468, para-norm 4879.314, grad-norm 2.714, time 7.07s, 
epoch 67, batches 6000|18183, train-acc 0.817, loss 0.464, para-norm 4878.892, grad-norm 5.023, time 6.93s, 
epoch 67, batches 7000|18183, train-acc 0.820, loss 0.456, para-norm 4878.346, grad-norm 1.844, time 6.92s, 
epoch 67, batches 8000|18183, train-acc 0.821, loss 0.463, para-norm 4877.203, grad-norm 13.195, time 6.93s, 
epoch 67, batches 9000|18183, train-acc 0.817, loss 0.462, para-norm 4877.091, grad-norm 2.747, time 6.99s, 
epoch 67, batches 10000|18183, train-acc 0.823, loss 0.453, para-norm 4875.657, grad-norm 5.835, time 6.93s, 
epoch 67, batches 11000|18183, train-acc 0.820, loss 0.455, para-norm 4876.031, grad-norm 1.960, time 6.92s, 
epoch 67, batches 12000|18183, train-acc 0.817, loss 0.467, para-norm 4875.209, grad-norm 1.306, time 6.92s, 
epoch 67, batches 13000|18183, train-acc 0.820, loss 0.456, para-norm 4873.668, grad-norm 3.044, time 7.04s, 
epoch 67, batches 14000|18183, train-acc 0.817, loss 0.466, para-norm 4873.095, grad-norm 3.760, time 7.74s, 
epoch 67, batches 15000|18183, train-acc 0.822, loss 0.452, para-norm 4872.656, grad-norm 4.067, time 7.74s, 
epoch 67, batches 16000|18183, train-acc 0.815, loss 0.468, para-norm 4872.436, grad-norm 2.800, time 7.74s, 
epoch 67, batches 17000|18183, train-acc 0.820, loss 0.459, para-norm 4871.478, grad-norm 1.460, time 7.57s, 
epoch 67, batches 18000|18183, train-acc 0.822, loss 0.455, para-norm 4871.742, grad-norm 1.957, time 7.02s, 
epoch 67, batches 18183|18183, train-acc 0.818, loss 0.455, para-norm 4871.315, grad-norm 0.902, time 1.27s, 
dev-acc 0.839
epoch 68, batches 1000|18183, train-acc 0.820, loss 0.457, para-norm 4872.819, grad-norm 2.722, time 6.99s, 
epoch 68, batches 2000|18183, train-acc 0.818, loss 0.461, para-norm 4871.920, grad-norm 4.273, time 6.94s, 
epoch 68, batches 3000|18183, train-acc 0.817, loss 0.460, para-norm 4872.087, grad-norm 5.811, time 6.93s, 
epoch 68, batches 4000|18183, train-acc 0.816, loss 0.465, para-norm 4871.498, grad-norm 5.112, time 6.93s, 
epoch 68, batches 5000|18183, train-acc 0.819, loss 0.452, para-norm 4871.600, grad-norm 2.178, time 6.93s, 
epoch 68, batches 6000|18183, train-acc 0.820, loss 0.460, para-norm 4871.138, grad-norm 3.482, time 7.04s, 
epoch 68, batches 7000|18183, train-acc 0.822, loss 0.451, para-norm 4870.603, grad-norm 2.529, time 6.94s, 
epoch 68, batches 8000|18183, train-acc 0.821, loss 0.455, para-norm 4871.061, grad-norm 9.039, time 6.94s, 
epoch 68, batches 9000|18183, train-acc 0.817, loss 0.465, para-norm 4869.994, grad-norm 1.959, time 6.93s, 
epoch 68, batches 10000|18183, train-acc 0.818, loss 0.459, para-norm 4869.470, grad-norm 5.397, time 6.92s, 
epoch 68, batches 11000|18183, train-acc 0.826, loss 0.451, para-norm 4869.564, grad-norm 4.286, time 6.95s, 
epoch 68, batches 12000|18183, train-acc 0.820, loss 0.459, para-norm 4869.512, grad-norm 4.308, time 6.94s, 
epoch 68, batches 13000|18183, train-acc 0.822, loss 0.458, para-norm 4868.326, grad-norm 3.979, time 6.95s, 
epoch 68, batches 14000|18183, train-acc 0.822, loss 0.457, para-norm 4867.751, grad-norm 1.979, time 6.94s, 
epoch 68, batches 15000|18183, train-acc 0.823, loss 0.451, para-norm 4867.775, grad-norm 2.282, time 6.91s, 
epoch 68, batches 16000|18183, train-acc 0.823, loss 0.448, para-norm 4868.827, grad-norm 1.919, time 6.92s, 
epoch 68, batches 17000|18183, train-acc 0.817, loss 0.462, para-norm 4868.416, grad-norm 7.509, time 6.92s, 
epoch 68, batches 18000|18183, train-acc 0.822, loss 0.453, para-norm 4866.911, grad-norm 2.358, time 6.94s, 
epoch 68, batches 18183|18183, train-acc 0.829, loss 0.443, para-norm 4867.339, grad-norm 2.514, time 1.27s, 
dev-acc 0.841
epoch 69, batches 1000|18183, train-acc 0.822, loss 0.450, para-norm 4868.665, grad-norm 4.878, time 7.02s, 
epoch 69, batches 2000|18183, train-acc 0.820, loss 0.448, para-norm 4868.053, grad-norm 1.904, time 6.93s, 
epoch 69, batches 3000|18183, train-acc 0.821, loss 0.455, para-norm 4867.970, grad-norm 1.100, time 6.95s, 
epoch 69, batches 4000|18183, train-acc 0.824, loss 0.449, para-norm 4876.843, grad-norm 2.138, time 6.94s, 
epoch 69, batches 5000|18183, train-acc 0.821, loss 0.455, para-norm 4876.550, grad-norm 2.570, time 6.95s, 
epoch 69, batches 6000|18183, train-acc 0.821, loss 0.457, para-norm 4874.871, grad-norm 3.138, time 6.94s, 
epoch 69, batches 7000|18183, train-acc 0.823, loss 0.453, para-norm 4873.685, grad-norm 4.831, time 6.95s, 
epoch 69, batches 8000|18183, train-acc 0.822, loss 0.454, para-norm 4873.911, grad-norm 2.803, time 6.93s, 
epoch 69, batches 9000|18183, train-acc 0.822, loss 0.448, para-norm 4874.195, grad-norm 0.441, time 6.94s, 
epoch 69, batches 10000|18183, train-acc 0.822, loss 0.453, para-norm 4872.742, grad-norm 4.621, time 6.92s, 
epoch 69, batches 11000|18183, train-acc 0.821, loss 0.456, para-norm 4871.839, grad-norm 3.383, time 6.92s, 
epoch 69, batches 12000|18183, train-acc 0.819, loss 0.459, para-norm 4879.601, grad-norm 3.840, time 6.94s, 
epoch 69, batches 13000|18183, train-acc 0.823, loss 0.449, para-norm 4879.363, grad-norm 5.549, time 6.95s, 
epoch 69, batches 14000|18183, train-acc 0.822, loss 0.455, para-norm 4879.272, grad-norm 1.066, time 6.92s, 
epoch 69, batches 15000|18183, train-acc 0.821, loss 0.457, para-norm 4880.601, grad-norm 4.614, time 6.93s, 
epoch 69, batches 16000|18183, train-acc 0.820, loss 0.457, para-norm 4879.732, grad-norm 2.972, time 6.93s, 
epoch 69, batches 17000|18183, train-acc 0.822, loss 0.453, para-norm 4878.846, grad-norm 1.312, time 6.91s, 
epoch 69, batches 18000|18183, train-acc 0.826, loss 0.445, para-norm 4877.946, grad-norm 3.349, time 6.95s, 
epoch 69, batches 18183|18183, train-acc 0.829, loss 0.436, para-norm 4878.299, grad-norm 1429.327, time 1.27s, 
dev-acc 0.839
epoch 70, batches 1000|18183, train-acc 0.825, loss 0.446, para-norm 4878.378, grad-norm 2.756, time 6.99s, 
epoch 70, batches 2000|18183, train-acc 0.828, loss 0.445, para-norm 4878.142, grad-norm 2.668, time 6.94s, 
epoch 70, batches 3000|18183, train-acc 0.821, loss 0.454, para-norm 4877.434, grad-norm 1.248, time 6.93s, 
epoch 70, batches 4000|18183, train-acc 0.826, loss 0.450, para-norm 4876.283, grad-norm 3.918, time 6.94s, 
epoch 70, batches 5000|18183, train-acc 0.822, loss 0.453, para-norm 4875.796, grad-norm 3.940, time 6.94s, 
epoch 70, batches 6000|18183, train-acc 0.822, loss 0.452, para-norm 4875.515, grad-norm 2.205, time 6.96s, 
epoch 70, batches 7000|18183, train-acc 0.822, loss 0.456, para-norm 4875.391, grad-norm 3.778, time 6.92s, 
epoch 70, batches 8000|18183, train-acc 0.823, loss 0.450, para-norm 4875.764, grad-norm 1.921, time 7.02s, 
epoch 70, batches 9000|18183, train-acc 0.824, loss 0.449, para-norm 4881.355, grad-norm 2.836, time 6.94s, 
epoch 70, batches 10000|18183, train-acc 0.824, loss 0.446, para-norm 4881.039, grad-norm 20.334, time 6.94s, 
epoch 70, batches 11000|18183, train-acc 0.819, loss 0.456, para-norm 4880.579, grad-norm 3.807, time 6.93s, 
epoch 70, batches 12000|18183, train-acc 0.821, loss 0.453, para-norm 4879.605, grad-norm 1.412, time 6.92s, 
epoch 70, batches 13000|18183, train-acc 0.819, loss 0.460, para-norm 4879.000, grad-norm 2.776, time 6.94s, 
epoch 70, batches 14000|18183, train-acc 0.825, loss 0.450, para-norm 4877.159, grad-norm 7.947, time 6.93s, 
epoch 70, batches 15000|18183, train-acc 0.826, loss 0.441, para-norm 4877.247, grad-norm 3.338, time 7.01s, 
epoch 70, batches 16000|18183, train-acc 0.824, loss 0.447, para-norm 4877.049, grad-norm 1.953, time 6.91s, 
epoch 70, batches 17000|18183, train-acc 0.825, loss 0.452, para-norm 4876.084, grad-norm 3.253, time 6.93s, 
epoch 70, batches 18000|18183, train-acc 0.823, loss 0.448, para-norm 4876.301, grad-norm 2.972, time 6.93s, 
epoch 70, batches 18183|18183, train-acc 0.815, loss 0.456, para-norm 4876.299, grad-norm 3.201, time 1.26s, 
dev-acc 0.840
epoch 71, batches 1000|18183, train-acc 0.826, loss 0.449, para-norm 4876.637, grad-norm 1.868, time 7.06s, 
epoch 71, batches 2000|18183, train-acc 0.827, loss 0.445, para-norm 4876.714, grad-norm 3.726, time 6.94s, 
epoch 71, batches 3000|18183, train-acc 0.828, loss 0.444, para-norm 4876.097, grad-norm 4.403, time 6.92s, 
epoch 71, batches 4000|18183, train-acc 0.824, loss 0.450, para-norm 4875.266, grad-norm 0.443, time 6.94s, 
epoch 71, batches 5000|18183, train-acc 0.824, loss 0.449, para-norm 4874.874, grad-norm 3.956, time 6.93s, 
epoch 71, batches 6000|18183, train-acc 0.826, loss 0.447, para-norm 4875.184, grad-norm 3.680, time 6.93s, 
epoch 71, batches 7000|18183, train-acc 0.822, loss 0.449, para-norm 4874.374, grad-norm 3.769, time 6.92s, 
epoch 71, batches 8000|18183, train-acc 0.827, loss 0.445, para-norm 4882.509, grad-norm 6.415, time 6.95s, 
epoch 71, batches 9000|18183, train-acc 0.824, loss 0.447, para-norm 4883.004, grad-norm 1.749, time 6.94s, 
epoch 71, batches 10000|18183, train-acc 0.824, loss 0.451, para-norm 4881.049, grad-norm 2.854, time 6.94s, 
epoch 71, batches 11000|18183, train-acc 0.824, loss 0.450, para-norm 4880.791, grad-norm 3.889, time 6.94s, 
epoch 71, batches 12000|18183, train-acc 0.824, loss 0.449, para-norm 4879.998, grad-norm 2.636, time 6.91s, 
epoch 71, batches 13000|18183, train-acc 0.830, loss 0.438, para-norm 4881.540, grad-norm 894.081, time 6.94s, 
epoch 71, batches 14000|18183, train-acc 0.823, loss 0.449, para-norm 4879.817, grad-norm 2.398, time 6.94s, 
epoch 71, batches 15000|18183, train-acc 0.822, loss 0.450, para-norm 4879.121, grad-norm 5.741, time 6.93s, 
epoch 71, batches 16000|18183, train-acc 0.823, loss 0.448, para-norm 4878.994, grad-norm 1.543, time 6.92s, 
epoch 71, batches 17000|18183, train-acc 0.825, loss 0.448, para-norm 4878.573, grad-norm 1.926, time 6.93s, 
epoch 71, batches 18000|18183, train-acc 0.825, loss 0.450, para-norm 4877.883, grad-norm 5.864, time 6.93s, 
epoch 71, batches 18183|18183, train-acc 0.822, loss 0.451, para-norm 4878.026, grad-norm 3.656, time 1.27s, 
dev-acc 0.840
epoch 72, batches 1000|18183, train-acc 0.828, loss 0.443, para-norm 4877.809, grad-norm 2.873, time 6.99s, 
epoch 72, batches 2000|18183, train-acc 0.828, loss 0.443, para-norm 4878.555, grad-norm 5.952, time 6.93s, 
epoch 72, batches 3000|18183, train-acc 0.824, loss 0.450, para-norm 4878.662, grad-norm 1.502, time 6.94s, 
epoch 72, batches 4000|18183, train-acc 0.825, loss 0.442, para-norm 4878.841, grad-norm 2.415, time 6.95s, 
epoch 72, batches 5000|18183, train-acc 0.829, loss 0.439, para-norm 4877.765, grad-norm 2.561, time 6.92s, 
epoch 72, batches 6000|18183, train-acc 0.827, loss 0.443, para-norm 4877.180, grad-norm 7.026, time 6.94s, 
epoch 72, batches 7000|18183, train-acc 0.825, loss 0.444, para-norm 4875.883, grad-norm 4.734, time 6.93s, 
epoch 72, batches 8000|18183, train-acc 0.826, loss 0.441, para-norm 4875.848, grad-norm 1.372, time 6.93s, 
epoch 72, batches 9000|18183, train-acc 0.823, loss 0.453, para-norm 4875.516, grad-norm 7.164, time 6.93s, 
epoch 72, batches 10000|18183, train-acc 0.824, loss 0.447, para-norm 4875.267, grad-norm 3.599, time 6.95s, 
epoch 72, batches 11000|18183, train-acc 0.828, loss 0.443, para-norm 4875.177, grad-norm 6.161, time 6.94s, 
epoch 72, batches 12000|18183, train-acc 0.821, loss 0.453, para-norm 4875.968, grad-norm 2.484, time 6.93s, 
epoch 72, batches 13000|18183, train-acc 0.822, loss 0.453, para-norm 4875.230, grad-norm 7.925, time 6.94s, 
epoch 72, batches 14000|18183, train-acc 0.827, loss 0.445, para-norm 4874.989, grad-norm 2.363, time 6.92s, 
epoch 72, batches 15000|18183, train-acc 0.827, loss 0.445, para-norm 4874.876, grad-norm 2.503, time 6.93s, 
epoch 72, batches 16000|18183, train-acc 0.828, loss 0.439, para-norm 4874.560, grad-norm 2.249, time 6.93s, 
epoch 72, batches 17000|18183, train-acc 0.826, loss 0.446, para-norm 4874.618, grad-norm 3.767, time 6.98s, 
epoch 72, batches 18000|18183, train-acc 0.824, loss 0.447, para-norm 4873.589, grad-norm 2.308, time 6.94s, 
epoch 72, batches 18183|18183, train-acc 0.820, loss 0.454, para-norm 4881.286, grad-norm 4.472, time 1.27s, 
dev-acc 0.841
epoch 73, batches 1000|18183, train-acc 0.830, loss 0.440, para-norm 4883.000, grad-norm 3.877, time 7.00s, 
epoch 73, batches 2000|18183, train-acc 0.830, loss 0.439, para-norm 4883.604, grad-norm 10.877, time 6.94s, 
epoch 73, batches 3000|18183, train-acc 0.826, loss 0.441, para-norm 4882.836, grad-norm 5.427, time 6.93s, 
epoch 73, batches 4000|18183, train-acc 0.825, loss 0.446, para-norm 4882.675, grad-norm 2.231, time 6.94s, 
epoch 73, batches 5000|18183, train-acc 0.824, loss 0.448, para-norm 4881.810, grad-norm 2.601, time 6.92s, 
epoch 73, batches 6000|18183, train-acc 0.830, loss 0.439, para-norm 4882.143, grad-norm 5.391, time 6.92s, 
epoch 73, batches 7000|18183, train-acc 0.830, loss 0.440, para-norm 4888.857, grad-norm 1.419, time 6.93s, 
epoch 73, batches 8000|18183, train-acc 0.825, loss 0.445, para-norm 4888.807, grad-norm 5.167, time 6.92s, 
epoch 73, batches 9000|18183, train-acc 0.829, loss 0.436, para-norm 4904.088, grad-norm 3.446, time 6.93s, 
epoch 73, batches 10000|18183, train-acc 0.823, loss 0.449, para-norm 4903.070, grad-norm 2.963, time 6.94s, 
epoch 73, batches 11000|18183, train-acc 0.823, loss 0.450, para-norm 4902.733, grad-norm 3.345, time 6.94s, 
epoch 73, batches 12000|18183, train-acc 0.827, loss 0.451, para-norm 4901.530, grad-norm 3.393, time 6.95s, 
epoch 73, batches 13000|18183, train-acc 0.827, loss 0.439, para-norm 4900.651, grad-norm 3.717, time 6.93s, 
epoch 73, batches 14000|18183, train-acc 0.823, loss 0.448, para-norm 4900.159, grad-norm 3.037, time 6.94s, 
epoch 73, batches 15000|18183, train-acc 0.827, loss 0.439, para-norm 4899.217, grad-norm 2.304, time 6.94s, 
epoch 73, batches 16000|18183, train-acc 0.825, loss 0.446, para-norm 4898.254, grad-norm 5.224, time 6.94s, 
epoch 73, batches 17000|18183, train-acc 0.828, loss 0.443, para-norm 4898.183, grad-norm 53.173, time 6.92s, 
epoch 73, batches 18000|18183, train-acc 0.830, loss 0.442, para-norm 4897.878, grad-norm 5.772, time 6.93s, 
epoch 73, batches 18183|18183, train-acc 0.826, loss 0.443, para-norm 4897.999, grad-norm 1.275, time 1.28s, 
dev-acc 0.845
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
save model!
epoch 74, batches 1000|18183, train-acc 0.827, loss 0.442, para-norm 4898.163, grad-norm 3.104, time 7.03s, 
epoch 74, batches 2000|18183, train-acc 0.830, loss 0.442, para-norm 4896.592, grad-norm 2.696, time 6.92s, 
epoch 74, batches 3000|18183, train-acc 0.825, loss 0.443, para-norm 4896.741, grad-norm 1.910, time 6.93s, 
epoch 74, batches 4000|18183, train-acc 0.829, loss 0.441, para-norm 4896.573, grad-norm 3.775, time 6.94s, 
epoch 74, batches 5000|18183, train-acc 0.829, loss 0.441, para-norm 4896.068, grad-norm 4.097, time 6.94s, 
epoch 74, batches 6000|18183, train-acc 0.827, loss 0.440, para-norm 4894.245, grad-norm 2.143, time 6.93s, 
epoch 74, batches 7000|18183, train-acc 0.827, loss 0.440, para-norm 4901.529, grad-norm 2.580, time 6.94s, 
epoch 74, batches 8000|18183, train-acc 0.827, loss 0.438, para-norm 4901.635, grad-norm 3.049, time 6.93s, 
epoch 74, batches 9000|18183, train-acc 0.828, loss 0.445, para-norm 4902.718, grad-norm 2.582, time 6.94s, 
epoch 74, batches 10000|18183, train-acc 0.831, loss 0.436, para-norm 4903.784, grad-norm 2.331, time 6.94s, 
epoch 74, batches 11000|18183, train-acc 0.827, loss 0.439, para-norm 4903.018, grad-norm 5.722, time 6.94s, 
epoch 74, batches 12000|18183, train-acc 0.832, loss 0.435, para-norm 4901.540, grad-norm 228.552, time 6.94s, 
epoch 74, batches 13000|18183, train-acc 0.825, loss 0.442, para-norm 4901.268, grad-norm 2.637, time 6.94s, 
epoch 74, batches 14000|18183, train-acc 0.827, loss 0.440, para-norm 4901.199, grad-norm 5.172, time 6.93s, 
epoch 74, batches 15000|18183, train-acc 0.823, loss 0.450, para-norm 4900.161, grad-norm 1.198, time 6.94s, 
epoch 74, batches 16000|18183, train-acc 0.824, loss 0.448, para-norm 4898.149, grad-norm 7.030, time 6.95s, 
epoch 74, batches 17000|18183, train-acc 0.826, loss 0.441, para-norm 4898.617, grad-norm 5.895, time 6.93s, 
epoch 74, batches 18000|18183, train-acc 0.829, loss 0.436, para-norm 4897.762, grad-norm 3.055, time 6.93s, 
epoch 74, batches 18183|18183, train-acc 0.831, loss 0.444, para-norm 4897.658, grad-norm 3.877, time 1.27s, 
dev-acc 0.842
epoch 75, batches 1000|18183, train-acc 0.828, loss 0.439, para-norm 4897.809, grad-norm 12.248, time 7.01s, 
epoch 75, batches 2000|18183, train-acc 0.830, loss 0.441, para-norm 4897.298, grad-norm 6.579, time 6.93s, 
epoch 75, batches 3000|18183, train-acc 0.829, loss 0.435, para-norm 4897.015, grad-norm 2.321, time 7.02s, 
epoch 75, batches 4000|18183, train-acc 0.828, loss 0.441, para-norm 4896.400, grad-norm 2.164, time 6.94s, 
epoch 75, batches 5000|18183, train-acc 0.827, loss 0.443, para-norm 4896.000, grad-norm 4.466, time 6.92s, 
epoch 75, batches 6000|18183, train-acc 0.832, loss 0.433, para-norm 4895.975, grad-norm 5.096, time 6.94s, 
epoch 75, batches 7000|18183, train-acc 0.828, loss 0.432, para-norm 4896.891, grad-norm 2.960, time 6.93s, 
epoch 75, batches 8000|18183, train-acc 0.832, loss 0.431, para-norm 4896.384, grad-norm 3.629, time 6.93s, 
epoch 75, batches 9000|18183, train-acc 0.827, loss 0.444, para-norm 4894.492, grad-norm 2.406, time 6.93s, 
epoch 75, batches 10000|18183, train-acc 0.830, loss 0.441, para-norm 4893.726, grad-norm 3.193, time 6.94s, 
epoch 75, batches 11000|18183, train-acc 0.827, loss 0.443, para-norm 4893.285, grad-norm 2.862, time 6.91s, 
epoch 75, batches 12000|18183, train-acc 0.829, loss 0.440, para-norm 4892.993, grad-norm 2.852, time 6.93s, 
epoch 75, batches 13000|18183, train-acc 0.828, loss 0.445, para-norm 4892.308, grad-norm 2.169, time 6.92s, 
epoch 75, batches 14000|18183, train-acc 0.827, loss 0.441, para-norm 4892.810, grad-norm 4.703, time 6.92s, 
epoch 75, batches 15000|18183, train-acc 0.828, loss 0.436, para-norm 4892.549, grad-norm 1.993, time 6.94s, 
epoch 75, batches 16000|18183, train-acc 0.829, loss 0.438, para-norm 4892.077, grad-norm 4.332, time 6.93s, 
epoch 75, batches 17000|18183, train-acc 0.828, loss 0.442, para-norm 4890.572, grad-norm 5.158, time 6.94s, 
epoch 75, batches 18000|18183, train-acc 0.829, loss 0.440, para-norm 4889.045, grad-norm 2.927, time 6.94s, 
epoch 75, batches 18183|18183, train-acc 0.830, loss 0.442, para-norm 4889.155, grad-norm 1.504, time 1.26s, 
dev-acc 0.846
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
save model!
epoch 76, batches 1000|18183, train-acc 0.831, loss 0.432, para-norm 4889.642, grad-norm 2.393, time 7.02s, 
epoch 76, batches 2000|18183, train-acc 0.827, loss 0.441, para-norm 4890.597, grad-norm 3.847, time 6.91s, 
epoch 76, batches 3000|18183, train-acc 0.831, loss 0.436, para-norm 4891.206, grad-norm 1.842, time 6.93s, 
epoch 76, batches 4000|18183, train-acc 0.829, loss 0.437, para-norm 4890.473, grad-norm 4.016, time 6.95s, 
epoch 76, batches 5000|18183, train-acc 0.832, loss 0.429, para-norm 4896.075, grad-norm 0.904, time 6.93s, 
epoch 76, batches 6000|18183, train-acc 0.830, loss 0.434, para-norm 4898.148, grad-norm 4.170, time 6.94s, 
epoch 76, batches 7000|18183, train-acc 0.829, loss 0.438, para-norm 4901.790, grad-norm 0.823, time 6.94s, 
epoch 76, batches 8000|18183, train-acc 0.830, loss 0.434, para-norm 4900.210, grad-norm 2.010, time 6.93s, 
epoch 76, batches 9000|18183, train-acc 0.829, loss 0.437, para-norm 4900.140, grad-norm 3.598, time 6.92s, 
epoch 76, batches 10000|18183, train-acc 0.826, loss 0.443, para-norm 4899.211, grad-norm 2.781, time 6.94s, 
epoch 76, batches 11000|18183, train-acc 0.831, loss 0.434, para-norm 4898.645, grad-norm 5.261, time 6.92s, 
epoch 76, batches 12000|18183, train-acc 0.828, loss 0.437, para-norm 4899.157, grad-norm 3429.407, time 6.93s, 
epoch 76, batches 13000|18183, train-acc 0.826, loss 0.444, para-norm 4898.623, grad-norm 4.347, time 6.93s, 
epoch 76, batches 14000|18183, train-acc 0.833, loss 0.432, para-norm 4900.182, grad-norm 2.007, time 6.93s, 
epoch 76, batches 15000|18183, train-acc 0.823, loss 0.451, para-norm 4898.921, grad-norm 4.639, time 6.94s, 
epoch 76, batches 16000|18183, train-acc 0.826, loss 0.447, para-norm 4897.895, grad-norm 3.113, time 6.93s, 
epoch 76, batches 17000|18183, train-acc 0.830, loss 0.434, para-norm 4897.174, grad-norm 3.635, time 6.92s, 
epoch 76, batches 18000|18183, train-acc 0.830, loss 0.437, para-norm 4896.269, grad-norm 4.378, time 6.93s, 
epoch 76, batches 18183|18183, train-acc 0.831, loss 0.432, para-norm 4896.176, grad-norm 11.140, time 1.27s, 
dev-acc 0.843
epoch 77, batches 1000|18183, train-acc 0.834, loss 0.428, para-norm 4896.520, grad-norm 409.473, time 6.96s, 
epoch 77, batches 2000|18183, train-acc 0.830, loss 0.435, para-norm 4896.107, grad-norm 5.641, time 6.94s, 
epoch 77, batches 3000|18183, train-acc 0.831, loss 0.436, para-norm 4896.902, grad-norm 4.074, time 6.93s, 
epoch 77, batches 4000|18183, train-acc 0.830, loss 0.434, para-norm 4895.948, grad-norm 6.798, time 6.92s, 
epoch 77, batches 5000|18183, train-acc 0.828, loss 0.438, para-norm 4896.326, grad-norm 5.267, time 6.93s, 
epoch 77, batches 6000|18183, train-acc 0.830, loss 0.431, para-norm 4896.983, grad-norm 6.410, time 6.93s, 
epoch 77, batches 7000|18183, train-acc 0.828, loss 0.438, para-norm 4897.574, grad-norm 4.031, time 6.94s, 
epoch 77, batches 8000|18183, train-acc 0.829, loss 0.438, para-norm 4897.283, grad-norm 10.135, time 6.92s, 
epoch 77, batches 9000|18183, train-acc 0.830, loss 0.435, para-norm 4895.657, grad-norm 3.362, time 6.94s, 
epoch 77, batches 10000|18183, train-acc 0.830, loss 0.437, para-norm 4895.102, grad-norm 2.135, time 6.94s, 
epoch 77, batches 11000|18183, train-acc 0.828, loss 0.436, para-norm 4894.767, grad-norm 5.409, time 6.92s, 
epoch 77, batches 12000|18183, train-acc 0.829, loss 0.440, para-norm 4893.794, grad-norm 3.715, time 6.92s, 
epoch 77, batches 13000|18183, train-acc 0.829, loss 0.438, para-norm 4893.392, grad-norm 1.350, time 6.94s, 
epoch 77, batches 14000|18183, train-acc 0.835, loss 0.429, para-norm 4893.186, grad-norm 14.454, time 7.07s, 
epoch 77, batches 15000|18183, train-acc 0.830, loss 0.434, para-norm 4892.833, grad-norm 2.014, time 7.08s, 
epoch 77, batches 16000|18183, train-acc 0.832, loss 0.439, para-norm 4893.190, grad-norm 3.709, time 6.94s, 
epoch 77, batches 17000|18183, train-acc 0.829, loss 0.439, para-norm 4891.222, grad-norm 3.717, time 6.93s, 
epoch 77, batches 18000|18183, train-acc 0.830, loss 0.438, para-norm 4891.309, grad-norm 70.540, time 6.92s, 
epoch 77, batches 18183|18183, train-acc 0.835, loss 0.428, para-norm 4891.444, grad-norm 6.244, time 1.26s, 
dev-acc 0.845
epoch 78, batches 1000|18183, train-acc 0.828, loss 0.438, para-norm 4891.381, grad-norm 1.669, time 7.71s, 
epoch 78, batches 2000|18183, train-acc 0.833, loss 0.431, para-norm 4890.839, grad-norm 4.229, time 7.74s, 
epoch 78, batches 3000|18183, train-acc 0.829, loss 0.435, para-norm 4891.659, grad-norm 3.793, time 7.77s, 
epoch 78, batches 4000|18183, train-acc 0.827, loss 0.437, para-norm 4891.111, grad-norm 1.751, time 7.76s, 
epoch 78, batches 5000|18183, train-acc 0.834, loss 0.430, para-norm 4890.942, grad-norm 3.695, time 7.76s, 
epoch 78, batches 6000|18183, train-acc 0.832, loss 0.432, para-norm 4890.359, grad-norm 3.094, time 7.74s, 
epoch 78, batches 7000|18183, train-acc 0.833, loss 0.426, para-norm 4891.689, grad-norm 5.281, time 7.76s, 
epoch 78, batches 8000|18183, train-acc 0.825, loss 0.442, para-norm 4890.841, grad-norm 1.411, time 7.78s, 
epoch 78, batches 9000|18183, train-acc 0.832, loss 0.431, para-norm 4889.715, grad-norm 2.062, time 7.74s, 
epoch 78, batches 10000|18183, train-acc 0.830, loss 0.434, para-norm 4889.882, grad-norm 4.976, time 7.75s, 
epoch 78, batches 11000|18183, train-acc 0.834, loss 0.431, para-norm 4888.933, grad-norm 35.741, time 7.76s, 
epoch 78, batches 12000|18183, train-acc 0.828, loss 0.440, para-norm 4887.973, grad-norm 2.846, time 7.76s, 
epoch 78, batches 13000|18183, train-acc 0.830, loss 0.438, para-norm 4887.280, grad-norm 2.935, time 7.77s, 
epoch 78, batches 14000|18183, train-acc 0.834, loss 0.429, para-norm 4886.414, grad-norm 2.418, time 7.76s, 
epoch 78, batches 15000|18183, train-acc 0.829, loss 0.443, para-norm 4885.933, grad-norm 1.216, time 7.75s, 
epoch 78, batches 16000|18183, train-acc 0.832, loss 0.436, para-norm 4885.540, grad-norm 2.688, time 7.75s, 
epoch 78, batches 17000|18183, train-acc 0.829, loss 0.437, para-norm 4885.079, grad-norm 1.380, time 7.75s, 
epoch 78, batches 18000|18183, train-acc 0.832, loss 0.433, para-norm 4885.285, grad-norm 5.387, time 7.76s, 
epoch 78, batches 18183|18183, train-acc 0.837, loss 0.417, para-norm 4885.887, grad-norm 2.650, time 1.42s, 
dev-acc 0.845
epoch 79, batches 1000|18183, train-acc 0.833, loss 0.432, para-norm 4886.109, grad-norm 5.184, time 7.84s, 
epoch 79, batches 2000|18183, train-acc 0.836, loss 0.423, para-norm 4886.437, grad-norm 5.356, time 7.75s, 
epoch 79, batches 3000|18183, train-acc 0.830, loss 0.435, para-norm 4890.964, grad-norm 1.898, time 7.76s, 
epoch 79, batches 4000|18183, train-acc 0.832, loss 0.425, para-norm 4890.593, grad-norm 4.022, time 7.75s, 
epoch 79, batches 5000|18183, train-acc 0.833, loss 0.432, para-norm 4890.961, grad-norm 2.439, time 7.75s, 
epoch 79, batches 6000|18183, train-acc 0.836, loss 0.425, para-norm 4891.632, grad-norm 3.469, time 7.74s, 
epoch 79, batches 7000|18183, train-acc 0.829, loss 0.432, para-norm 4891.687, grad-norm 5.638, time 7.76s, 
epoch 79, batches 8000|18183, train-acc 0.831, loss 0.430, para-norm 4891.327, grad-norm 4.821, time 7.75s, 
epoch 79, batches 9000|18183, train-acc 0.830, loss 0.434, para-norm 4891.051, grad-norm 5.602, time 7.78s, 
epoch 79, batches 10000|18183, train-acc 0.831, loss 0.433, para-norm 4889.808, grad-norm 1.428, time 7.77s, 
epoch 79, batches 11000|18183, train-acc 0.828, loss 0.438, para-norm 4889.373, grad-norm 3.756, time 7.75s, 
epoch 79, batches 12000|18183, train-acc 0.831, loss 0.436, para-norm 4889.076, grad-norm 9.211, time 7.76s, 
epoch 79, batches 13000|18183, train-acc 0.827, loss 0.442, para-norm 4889.665, grad-norm 2.913, time 7.75s, 
epoch 79, batches 14000|18183, train-acc 0.832, loss 0.433, para-norm 4888.749, grad-norm 3.259, time 7.75s, 
epoch 79, batches 15000|18183, train-acc 0.830, loss 0.434, para-norm 4888.787, grad-norm 3.310, time 7.75s, 
epoch 79, batches 16000|18183, train-acc 0.833, loss 0.427, para-norm 4888.235, grad-norm 1.736, time 7.77s, 
epoch 79, batches 17000|18183, train-acc 0.829, loss 0.439, para-norm 4888.075, grad-norm 2.918, time 7.76s, 
epoch 79, batches 18000|18183, train-acc 0.832, loss 0.436, para-norm 4886.977, grad-norm 2.916, time 7.75s, 
epoch 79, batches 18183|18183, train-acc 0.831, loss 0.423, para-norm 4886.920, grad-norm 3.019, time 1.42s, 
dev-acc 0.847
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
save model!
epoch 80, batches 1000|18183, train-acc 0.831, loss 0.431, para-norm 4887.304, grad-norm 3.125, time 7.32s, 
epoch 80, batches 2000|18183, train-acc 0.830, loss 0.435, para-norm 4887.180, grad-norm 3.273, time 6.95s, 
epoch 80, batches 3000|18183, train-acc 0.830, loss 0.431, para-norm 4888.817, grad-norm 3.194, time 6.94s, 
epoch 80, batches 4000|18183, train-acc 0.832, loss 0.434, para-norm 4888.484, grad-norm 3.378, time 6.94s, 
epoch 80, batches 5000|18183, train-acc 0.834, loss 0.427, para-norm 4889.760, grad-norm 1.852, time 6.93s, 
epoch 80, batches 6000|18183, train-acc 0.830, loss 0.437, para-norm 4890.289, grad-norm 4.753, time 6.95s, 
epoch 80, batches 7000|18183, train-acc 0.833, loss 0.433, para-norm 4890.241, grad-norm 2.654, time 6.95s, 
epoch 80, batches 8000|18183, train-acc 0.834, loss 0.430, para-norm 4889.125, grad-norm 2.271, time 6.95s, 
epoch 80, batches 9000|18183, train-acc 0.831, loss 0.431, para-norm 4889.196, grad-norm 1.565, time 6.94s, 
epoch 80, batches 10000|18183, train-acc 0.836, loss 0.424, para-norm 4889.697, grad-norm 21.597, time 6.94s, 
epoch 80, batches 11000|18183, train-acc 0.832, loss 0.430, para-norm 4889.252, grad-norm 4.714, time 6.93s, 
epoch 80, batches 12000|18183, train-acc 0.833, loss 0.431, para-norm 4895.919, grad-norm 4.028, time 7.08s, 
epoch 80, batches 13000|18183, train-acc 0.832, loss 0.430, para-norm 4895.592, grad-norm 2.707, time 6.94s, 
epoch 80, batches 14000|18183, train-acc 0.833, loss 0.428, para-norm 4895.331, grad-norm 4.319, time 6.95s, 
epoch 80, batches 15000|18183, train-acc 0.830, loss 0.438, para-norm 4895.817, grad-norm 4.939, time 6.94s, 
epoch 80, batches 16000|18183, train-acc 0.838, loss 0.422, para-norm 4895.183, grad-norm 2.402, time 6.93s, 
epoch 80, batches 17000|18183, train-acc 0.831, loss 0.431, para-norm 4901.661, grad-norm 3.018, time 6.95s, 
epoch 80, batches 18000|18183, train-acc 0.832, loss 0.430, para-norm 4901.667, grad-norm 2.446, time 6.93s, 
epoch 80, batches 18183|18183, train-acc 0.833, loss 0.429, para-norm 4901.662, grad-norm 4.294, time 1.27s, 
dev-acc 0.848
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
save model!
epoch 81, batches 1000|18183, train-acc 0.835, loss 0.425, para-norm 4910.976, grad-norm 5.208, time 7.05s, 
epoch 81, batches 2000|18183, train-acc 0.838, loss 0.419, para-norm 4911.440, grad-norm 1.269, time 6.94s, 
epoch 81, batches 3000|18183, train-acc 0.832, loss 0.433, para-norm 4911.040, grad-norm 2.156, time 6.95s, 
epoch 81, batches 4000|18183, train-acc 0.831, loss 0.432, para-norm 4910.398, grad-norm 2.315, time 6.94s, 
epoch 81, batches 5000|18183, train-acc 0.837, loss 0.424, para-norm 4909.166, grad-norm 4.484, time 6.92s, 
epoch 81, batches 6000|18183, train-acc 0.835, loss 0.428, para-norm 4909.086, grad-norm 16.981, time 6.94s, 
epoch 81, batches 7000|18183, train-acc 0.833, loss 0.430, para-norm 4908.068, grad-norm 2.652, time 6.93s, 
epoch 81, batches 8000|18183, train-acc 0.832, loss 0.429, para-norm 4913.846, grad-norm 1.574, time 6.94s, 
epoch 81, batches 9000|18183, train-acc 0.833, loss 0.430, para-norm 4913.715, grad-norm 8.326, time 6.95s, 
epoch 81, batches 10000|18183, train-acc 0.832, loss 0.431, para-norm 4913.931, grad-norm 3.383, time 6.93s, 
epoch 81, batches 11000|18183, train-acc 0.833, loss 0.430, para-norm 4914.191, grad-norm 1.208, time 6.95s, 
epoch 81, batches 12000|18183, train-acc 0.828, loss 0.438, para-norm 4912.904, grad-norm 4.595, time 6.94s, 
epoch 81, batches 13000|18183, train-acc 0.836, loss 0.425, para-norm 4912.964, grad-norm 5.140, time 6.94s, 
epoch 81, batches 14000|18183, train-acc 0.829, loss 0.438, para-norm 4912.442, grad-norm 8.499, time 6.93s, 
epoch 81, batches 15000|18183, train-acc 0.834, loss 0.423, para-norm 4912.185, grad-norm 2.803, time 6.93s, 
epoch 81, batches 16000|18183, train-acc 0.833, loss 0.427, para-norm 4911.615, grad-norm 2.743, time 6.94s, 
epoch 81, batches 17000|18183, train-acc 0.830, loss 0.432, para-norm 4911.809, grad-norm 2.359, time 6.95s, 
epoch 81, batches 18000|18183, train-acc 0.834, loss 0.424, para-norm 4911.468, grad-norm 4.366, time 6.94s, 
epoch 81, batches 18183|18183, train-acc 0.833, loss 0.420, para-norm 4911.518, grad-norm 0.999, time 1.27s, 
dev-acc 0.849
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
save model!
epoch 82, batches 1000|18183, train-acc 0.833, loss 0.429, para-norm 4910.994, grad-norm 4.770, time 7.03s, 
epoch 82, batches 2000|18183, train-acc 0.834, loss 0.424, para-norm 4911.982, grad-norm 1.786, time 6.95s, 
epoch 82, batches 3000|18183, train-acc 0.833, loss 0.424, para-norm 4911.575, grad-norm 4.583, time 6.93s, 
epoch 82, batches 4000|18183, train-acc 0.834, loss 0.427, para-norm 4911.290, grad-norm 3.165, time 6.94s, 
epoch 82, batches 5000|18183, train-acc 0.831, loss 0.437, para-norm 4911.258, grad-norm 1.630, time 6.96s, 
epoch 82, batches 6000|18183, train-acc 0.833, loss 0.427, para-norm 4910.655, grad-norm 6.144, time 6.94s, 
epoch 82, batches 7000|18183, train-acc 0.831, loss 0.431, para-norm 4910.070, grad-norm 5.550, time 6.94s, 
epoch 82, batches 8000|18183, train-acc 0.836, loss 0.427, para-norm 4910.552, grad-norm 2.527, time 6.95s, 
epoch 82, batches 9000|18183, train-acc 0.838, loss 0.421, para-norm 4915.092, grad-norm 3.121, time 6.93s, 
epoch 82, batches 10000|18183, train-acc 0.838, loss 0.422, para-norm 4914.855, grad-norm 1.741, time 6.93s, 
epoch 82, batches 11000|18183, train-acc 0.835, loss 0.422, para-norm 4914.159, grad-norm 1.211, time 6.94s, 
epoch 82, batches 12000|18183, train-acc 0.834, loss 0.429, para-norm 4913.202, grad-norm 3.710, time 6.95s, 
epoch 82, batches 13000|18183, train-acc 0.834, loss 0.428, para-norm 4913.539, grad-norm 1.895, time 6.96s, 
epoch 82, batches 14000|18183, train-acc 0.832, loss 0.433, para-norm 4913.007, grad-norm 3.447, time 6.93s, 
epoch 82, batches 15000|18183, train-acc 0.837, loss 0.424, para-norm 4913.755, grad-norm 2.495, time 6.93s, 
epoch 82, batches 16000|18183, train-acc 0.838, loss 0.423, para-norm 4912.777, grad-norm 5.208, time 6.92s, 
epoch 82, batches 17000|18183, train-acc 0.833, loss 0.429, para-norm 4912.066, grad-norm 1.959, time 6.94s, 
epoch 82, batches 18000|18183, train-acc 0.833, loss 0.430, para-norm 4912.119, grad-norm 3.054, time 7.22s, 
epoch 82, batches 18183|18183, train-acc 0.826, loss 0.445, para-norm 4911.481, grad-norm 1.526, time 1.32s, 
dev-acc 0.845
epoch 83, batches 1000|18183, train-acc 0.835, loss 0.422, para-norm 4912.244, grad-norm 2.316, time 6.97s, 
epoch 83, batches 2000|18183, train-acc 0.836, loss 0.425, para-norm 4911.159, grad-norm 2.845, time 7.02s, 
epoch 83, batches 3000|18183, train-acc 0.835, loss 0.425, para-norm 4911.571, grad-norm 3.079, time 6.93s, 
epoch 83, batches 4000|18183, train-acc 0.833, loss 0.430, para-norm 4910.970, grad-norm 1.903, time 6.96s, 
epoch 83, batches 5000|18183, train-acc 0.836, loss 0.423, para-norm 4911.372, grad-norm 3.977, time 6.95s, 
epoch 83, batches 6000|18183, train-acc 0.836, loss 0.425, para-norm 4911.827, grad-norm 3.184, time 6.93s, 
epoch 83, batches 7000|18183, train-acc 0.834, loss 0.421, para-norm 4911.800, grad-norm 2.931, time 6.93s, 
epoch 83, batches 8000|18183, train-acc 0.836, loss 0.427, para-norm 4915.075, grad-norm 3.153, time 6.94s, 
epoch 83, batches 9000|18183, train-acc 0.835, loss 0.429, para-norm 4913.798, grad-norm 1.039, time 6.94s, 
epoch 83, batches 10000|18183, train-acc 0.836, loss 0.426, para-norm 4913.981, grad-norm 2.799, time 6.93s, 
epoch 83, batches 11000|18183, train-acc 0.832, loss 0.429, para-norm 4919.116, grad-norm 1318.245, time 6.92s, 
epoch 83, batches 12000|18183, train-acc 0.836, loss 0.422, para-norm 4919.274, grad-norm 2.716, time 6.95s, 
epoch 83, batches 13000|18183, train-acc 0.835, loss 0.426, para-norm 4918.823, grad-norm 2.484, time 6.94s, 
epoch 83, batches 14000|18183, train-acc 0.840, loss 0.417, para-norm 4919.014, grad-norm 2.945, time 6.94s, 
epoch 83, batches 15000|18183, train-acc 0.833, loss 0.429, para-norm 4920.004, grad-norm 0.111, time 6.95s, 
epoch 83, batches 16000|18183, train-acc 0.832, loss 0.434, para-norm 4918.397, grad-norm 1.124, time 6.96s, 
epoch 83, batches 17000|18183, train-acc 0.837, loss 0.426, para-norm 4918.056, grad-norm 3.871, time 7.01s, 
epoch 83, batches 18000|18183, train-acc 0.834, loss 0.423, para-norm 4917.510, grad-norm 5.302, time 6.95s, 
epoch 83, batches 18183|18183, train-acc 0.837, loss 0.428, para-norm 4917.283, grad-norm 2.422, time 1.27s, 
dev-acc 0.850
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
save model!
epoch 84, batches 1000|18183, train-acc 0.834, loss 0.425, para-norm 4924.587, grad-norm 2.189, time 6.99s, 
epoch 84, batches 2000|18183, train-acc 0.836, loss 0.421, para-norm 4925.913, grad-norm 3.818, time 6.93s, 
epoch 84, batches 3000|18183, train-acc 0.837, loss 0.419, para-norm 4926.452, grad-norm 6.790, time 6.94s, 
epoch 84, batches 4000|18183, train-acc 0.832, loss 0.432, para-norm 4926.000, grad-norm 3.785, time 6.95s, 
epoch 84, batches 5000|18183, train-acc 0.834, loss 0.429, para-norm 4925.472, grad-norm 6.976, time 6.95s, 
epoch 84, batches 6000|18183, train-acc 0.838, loss 0.419, para-norm 4926.147, grad-norm 4.643, time 6.94s, 
epoch 84, batches 7000|18183, train-acc 0.832, loss 0.427, para-norm 4927.407, grad-norm 5.132, time 6.96s, 
epoch 84, batches 8000|18183, train-acc 0.832, loss 0.426, para-norm 4926.729, grad-norm 3.227, time 6.95s, 
epoch 84, batches 9000|18183, train-acc 0.836, loss 0.423, para-norm 4927.243, grad-norm 3.367, time 6.92s, 
epoch 84, batches 10000|18183, train-acc 0.835, loss 0.423, para-norm 4926.345, grad-norm 4.541, time 6.95s, 
epoch 84, batches 11000|18183, train-acc 0.832, loss 0.432, para-norm 4926.538, grad-norm 4.787, time 6.95s, 
epoch 84, batches 12000|18183, train-acc 0.833, loss 0.430, para-norm 4925.501, grad-norm 3.691, time 6.94s, 
epoch 84, batches 13000|18183, train-acc 0.839, loss 0.416, para-norm 4925.353, grad-norm 2.908, time 6.94s, 
epoch 84, batches 14000|18183, train-acc 0.834, loss 0.426, para-norm 4924.792, grad-norm 4.352, time 6.93s, 
epoch 84, batches 15000|18183, train-acc 0.835, loss 0.427, para-norm 4923.671, grad-norm 4.986, time 6.96s, 
epoch 84, batches 16000|18183, train-acc 0.838, loss 0.419, para-norm 4926.371, grad-norm 1.715, time 6.94s, 
epoch 84, batches 17000|18183, train-acc 0.837, loss 0.424, para-norm 4926.038, grad-norm 6.162, time 6.94s, 
epoch 84, batches 18000|18183, train-acc 0.838, loss 0.424, para-norm 4924.223, grad-norm 11.008, time 6.95s, 
epoch 84, batches 18183|18183, train-acc 0.839, loss 0.416, para-norm 4924.471, grad-norm 3.554, time 1.28s, 
dev-acc 0.849
epoch 85, batches 1000|18183, train-acc 0.836, loss 0.423, para-norm 4924.488, grad-norm 91.181, time 7.04s, 
epoch 85, batches 2000|18183, train-acc 0.834, loss 0.422, para-norm 4925.828, grad-norm 2.117, time 6.95s, 
epoch 85, batches 3000|18183, train-acc 0.834, loss 0.426, para-norm 4925.332, grad-norm 4.664, time 6.95s, 
epoch 85, batches 4000|18183, train-acc 0.841, loss 0.414, para-norm 4925.176, grad-norm 4.374, time 6.92s, 
epoch 85, batches 5000|18183, train-acc 0.835, loss 0.422, para-norm 4930.891, grad-norm 2.674, time 6.95s, 
epoch 85, batches 6000|18183, train-acc 0.838, loss 0.420, para-norm 4930.367, grad-norm 4.417, time 6.95s, 
epoch 85, batches 7000|18183, train-acc 0.835, loss 0.425, para-norm 4931.177, grad-norm 2.136, time 6.95s, 
epoch 85, batches 8000|18183, train-acc 0.834, loss 0.421, para-norm 4930.423, grad-norm 3.825, time 6.95s, 
epoch 85, batches 9000|18183, train-acc 0.839, loss 0.416, para-norm 4930.046, grad-norm 5.378, time 6.93s, 
epoch 85, batches 10000|18183, train-acc 0.834, loss 0.429, para-norm 4928.728, grad-norm 2.467, time 6.94s, 
epoch 85, batches 11000|18183, train-acc 0.836, loss 0.421, para-norm 4928.697, grad-norm 2.027, time 6.95s, 
epoch 85, batches 12000|18183, train-acc 0.832, loss 0.433, para-norm 4927.679, grad-norm 2.866, time 6.95s, 
epoch 85, batches 13000|18183, train-acc 0.838, loss 0.425, para-norm 4927.235, grad-norm 5.745, time 6.94s, 
epoch 85, batches 14000|18183, train-acc 0.833, loss 0.426, para-norm 4927.296, grad-norm 1.715, time 7.09s, 
epoch 85, batches 15000|18183, train-acc 0.837, loss 0.421, para-norm 4928.145, grad-norm 3.780, time 7.19s, 
epoch 85, batches 16000|18183, train-acc 0.835, loss 0.426, para-norm 4928.506, grad-norm 2.349, time 7.19s, 
epoch 85, batches 17000|18183, train-acc 0.835, loss 0.423, para-norm 4927.291, grad-norm 4.517, time 7.20s, 
epoch 85, batches 18000|18183, train-acc 0.834, loss 0.428, para-norm 4926.780, grad-norm 3.222, time 7.18s, 
epoch 85, batches 18183|18183, train-acc 0.841, loss 0.409, para-norm 4927.176, grad-norm 2.070, time 1.31s, 
dev-acc 0.849
epoch 86, batches 1000|18183, train-acc 0.842, loss 0.411, para-norm 4926.539, grad-norm 2.286, time 7.27s, 
epoch 86, batches 2000|18183, train-acc 0.836, loss 0.419, para-norm 4926.814, grad-norm 2.319, time 7.20s, 
epoch 86, batches 3000|18183, train-acc 0.841, loss 0.415, para-norm 4927.414, grad-norm 2.946, time 7.20s, 
epoch 86, batches 4000|18183, train-acc 0.832, loss 0.425, para-norm 4927.018, grad-norm 4.800, time 7.20s, 
epoch 86, batches 5000|18183, train-acc 0.840, loss 0.413, para-norm 4926.599, grad-norm 3.047, time 7.20s, 
epoch 86, batches 6000|18183, train-acc 0.837, loss 0.424, para-norm 4927.302, grad-norm 4.158, time 7.20s, 
epoch 86, batches 7000|18183, train-acc 0.836, loss 0.419, para-norm 4927.922, grad-norm 2.776, time 7.20s, 
epoch 86, batches 8000|18183, train-acc 0.837, loss 0.421, para-norm 4927.191, grad-norm 2.409, time 7.20s, 
epoch 86, batches 9000|18183, train-acc 0.837, loss 0.418, para-norm 4927.314, grad-norm 6.634, time 7.21s, 
epoch 86, batches 10000|18183, train-acc 0.839, loss 0.417, para-norm 4933.615, grad-norm 366.556, time 7.20s, 
epoch 86, batches 11000|18183, train-acc 0.830, loss 0.433, para-norm 4933.595, grad-norm 7.659, time 7.20s, 
epoch 86, batches 12000|18183, train-acc 0.835, loss 0.422, para-norm 4933.907, grad-norm 3.560, time 7.21s, 
epoch 86, batches 13000|18183, train-acc 0.837, loss 0.424, para-norm 4939.946, grad-norm 1.457, time 7.20s, 
epoch 86, batches 14000|18183, train-acc 0.836, loss 0.423, para-norm 4939.620, grad-norm 2.566, time 7.20s, 
epoch 86, batches 15000|18183, train-acc 0.835, loss 0.423, para-norm 4939.820, grad-norm 3.057, time 7.20s, 
epoch 86, batches 16000|18183, train-acc 0.834, loss 0.430, para-norm 4940.019, grad-norm 7.034, time 7.20s, 
epoch 86, batches 17000|18183, train-acc 0.832, loss 0.432, para-norm 4939.849, grad-norm 5.254, time 7.20s, 
epoch 86, batches 18000|18183, train-acc 0.837, loss 0.422, para-norm 4939.161, grad-norm 4.986, time 7.20s, 
epoch 86, batches 18183|18183, train-acc 0.829, loss 0.424, para-norm 4938.631, grad-norm 5.391, time 1.32s, 
dev-acc 0.848
epoch 87, batches 1000|18183, train-acc 0.835, loss 0.422, para-norm 4939.041, grad-norm 3.862, time 7.28s, 
epoch 87, batches 2000|18183, train-acc 0.840, loss 0.411, para-norm 4939.213, grad-norm 0.284, time 7.19s, 
epoch 87, batches 3000|18183, train-acc 0.837, loss 0.421, para-norm 4938.804, grad-norm 4.535, time 7.22s, 
epoch 87, batches 4000|18183, train-acc 0.834, loss 0.425, para-norm 4939.280, grad-norm 3.987, time 7.22s, 
epoch 87, batches 5000|18183, train-acc 0.837, loss 0.419, para-norm 4938.883, grad-norm 2.888, time 7.21s, 
epoch 87, batches 6000|18183, train-acc 0.836, loss 0.423, para-norm 4939.834, grad-norm 6.358, time 7.21s, 
epoch 87, batches 7000|18183, train-acc 0.834, loss 0.425, para-norm 4940.579, grad-norm 2.388, time 7.21s, 
epoch 87, batches 8000|18183, train-acc 0.837, loss 0.420, para-norm 4939.887, grad-norm 2.432, time 7.20s, 
epoch 87, batches 9000|18183, train-acc 0.834, loss 0.422, para-norm 4939.766, grad-norm 2.480, time 7.20s, 
epoch 87, batches 10000|18183, train-acc 0.835, loss 0.424, para-norm 4939.190, grad-norm 0.003, time 7.20s, 
epoch 87, batches 11000|18183, train-acc 0.836, loss 0.420, para-norm 4938.872, grad-norm 4.150, time 7.20s, 
epoch 87, batches 12000|18183, train-acc 0.840, loss 0.416, para-norm 4939.144, grad-norm 0.766, time 7.21s, 
epoch 87, batches 13000|18183, train-acc 0.837, loss 0.421, para-norm 4938.776, grad-norm 2.198, time 7.17s, 
epoch 87, batches 14000|18183, train-acc 0.838, loss 0.412, para-norm 4939.746, grad-norm 3.609, time 7.11s, 
epoch 87, batches 15000|18183, train-acc 0.837, loss 0.422, para-norm 4939.224, grad-norm 5.944, time 6.94s, 
epoch 87, batches 16000|18183, train-acc 0.833, loss 0.424, para-norm 4938.806, grad-norm 5.784, time 7.03s, 
epoch 87, batches 17000|18183, train-acc 0.834, loss 0.424, para-norm 4938.463, grad-norm 3.261, time 6.95s, 
epoch 87, batches 18000|18183, train-acc 0.836, loss 0.426, para-norm 4937.743, grad-norm 5.034, time 6.95s, 
epoch 87, batches 18183|18183, train-acc 0.842, loss 0.415, para-norm 4937.281, grad-norm 3.068, time 1.27s, 
dev-acc 0.848
epoch 88, batches 1000|18183, train-acc 0.835, loss 0.424, para-norm 4937.119, grad-norm 18.185, time 7.04s, 
epoch 88, batches 2000|18183, train-acc 0.839, loss 0.409, para-norm 4937.961, grad-norm 8.434, time 6.95s, 
epoch 88, batches 3000|18183, train-acc 0.841, loss 0.409, para-norm 4938.582, grad-norm 2.885, time 6.95s, 
epoch 88, batches 4000|18183, train-acc 0.840, loss 0.416, para-norm 4938.858, grad-norm 433.820, time 6.95s, 
epoch 88, batches 5000|18183, train-acc 0.834, loss 0.426, para-norm 4938.768, grad-norm 4.813, time 6.94s, 
epoch 88, batches 6000|18183, train-acc 0.841, loss 0.414, para-norm 4938.656, grad-norm 3.378, time 6.95s, 
epoch 88, batches 7000|18183, train-acc 0.837, loss 0.419, para-norm 4938.655, grad-norm 2.474, time 6.95s, 
epoch 88, batches 8000|18183, train-acc 0.832, loss 0.428, para-norm 4938.069, grad-norm 3.237, time 6.95s, 
epoch 88, batches 9000|18183, train-acc 0.836, loss 0.425, para-norm 4938.981, grad-norm 10.859, time 6.95s, 
epoch 88, batches 10000|18183, train-acc 0.841, loss 0.413, para-norm 4939.672, grad-norm 3.412, time 6.94s, 
epoch 88, batches 11000|18183, train-acc 0.839, loss 0.416, para-norm 4945.979, grad-norm 3.281, time 6.94s, 
epoch 88, batches 12000|18183, train-acc 0.839, loss 0.418, para-norm 4945.866, grad-norm 3.114, time 6.95s, 
epoch 88, batches 13000|18183, train-acc 0.836, loss 0.422, para-norm 4945.578, grad-norm 3.592, time 6.94s, 
epoch 88, batches 14000|18183, train-acc 0.838, loss 0.420, para-norm 4952.434, grad-norm 2.667, time 6.95s, 
epoch 88, batches 15000|18183, train-acc 0.840, loss 0.418, para-norm 4961.530, grad-norm 2.625, time 6.96s, 
epoch 88, batches 16000|18183, train-acc 0.837, loss 0.418, para-norm 4960.121, grad-norm 1.736, time 6.94s, 
epoch 88, batches 17000|18183, train-acc 0.831, loss 0.431, para-norm 4958.387, grad-norm 7.211, time 6.95s, 
epoch 88, batches 18000|18183, train-acc 0.837, loss 0.418, para-norm 4958.642, grad-norm 4.799, time 6.94s, 
epoch 88, batches 18183|18183, train-acc 0.842, loss 0.408, para-norm 4958.565, grad-norm 965.245, time 1.27s, 
dev-acc 0.853
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
save model!
epoch 89, batches 1000|18183, train-acc 0.845, loss 0.407, para-norm 4959.521, grad-norm 4.367, time 7.02s, 
epoch 89, batches 2000|18183, train-acc 0.838, loss 0.418, para-norm 4958.789, grad-norm 2.046, time 6.95s, 
epoch 89, batches 3000|18183, train-acc 0.835, loss 0.421, para-norm 4960.113, grad-norm 0.017, time 6.95s, 
epoch 89, batches 4000|18183, train-acc 0.836, loss 0.423, para-norm 4959.913, grad-norm 4.424, time 6.93s, 
epoch 89, batches 5000|18183, train-acc 0.834, loss 0.426, para-norm 4960.324, grad-norm 0.940, time 6.92s, 
epoch 89, batches 6000|18183, train-acc 0.838, loss 0.420, para-norm 4960.296, grad-norm 2.761, time 6.95s, 
epoch 89, batches 7000|18183, train-acc 0.838, loss 0.415, para-norm 4960.134, grad-norm 7.934, time 6.94s, 
epoch 89, batches 8000|18183, train-acc 0.840, loss 0.412, para-norm 4960.295, grad-norm 1.808, time 7.02s, 
epoch 89, batches 9000|18183, train-acc 0.839, loss 0.414, para-norm 4958.634, grad-norm 3.859, time 6.94s, 
epoch 89, batches 10000|18183, train-acc 0.838, loss 0.414, para-norm 4958.859, grad-norm 2.793, time 7.04s, 
epoch 89, batches 11000|18183, train-acc 0.834, loss 0.423, para-norm 4957.992, grad-norm 0.333, time 6.95s, 
epoch 89, batches 12000|18183, train-acc 0.835, loss 0.423, para-norm 4957.164, grad-norm 0.244, time 6.96s, 
epoch 89, batches 13000|18183, train-acc 0.832, loss 0.423, para-norm 4955.855, grad-norm 4.830, time 6.94s, 
epoch 89, batches 14000|18183, train-acc 0.837, loss 0.422, para-norm 4955.043, grad-norm 3.028, time 6.95s, 
epoch 89, batches 15000|18183, train-acc 0.836, loss 0.422, para-norm 4954.253, grad-norm 5.337, time 6.95s, 
epoch 89, batches 16000|18183, train-acc 0.841, loss 0.417, para-norm 4953.366, grad-norm 5.098, time 6.95s, 
epoch 89, batches 17000|18183, train-acc 0.837, loss 0.418, para-norm 4953.113, grad-norm 1.642, time 6.95s, 
epoch 89, batches 18000|18183, train-acc 0.838, loss 0.418, para-norm 4952.532, grad-norm 3.433, time 6.94s, 
epoch 89, batches 18183|18183, train-acc 0.831, loss 0.426, para-norm 4952.777, grad-norm 4.055, time 1.27s, 
dev-acc 0.853
epoch 90, batches 1000|18183, train-acc 0.842, loss 0.414, para-norm 4952.813, grad-norm 2.356, time 6.97s, 
epoch 90, batches 2000|18183, train-acc 0.839, loss 0.417, para-norm 4952.600, grad-norm 4.160, time 6.95s, 
epoch 90, batches 3000|18183, train-acc 0.836, loss 0.423, para-norm 4952.754, grad-norm 5.146, time 6.96s, 
epoch 90, batches 4000|18183, train-acc 0.838, loss 0.417, para-norm 4953.245, grad-norm 4.057, time 6.95s, 
epoch 90, batches 5000|18183, train-acc 0.835, loss 0.422, para-norm 4953.307, grad-norm 4.566, time 6.95s, 
epoch 90, batches 6000|18183, train-acc 0.834, loss 0.426, para-norm 4952.688, grad-norm 8.310, time 6.94s, 
epoch 90, batches 7000|18183, train-acc 0.840, loss 0.414, para-norm 4952.758, grad-norm 2.384, time 6.93s, 
epoch 90, batches 8000|18183, train-acc 0.838, loss 0.418, para-norm 4951.842, grad-norm 4.468, time 6.95s, 
epoch 90, batches 9000|18183, train-acc 0.841, loss 0.411, para-norm 4951.748, grad-norm 6.371, time 6.95s, 
epoch 90, batches 10000|18183, train-acc 0.840, loss 0.408, para-norm 4951.811, grad-norm 6.065, time 6.94s, 
epoch 90, batches 11000|18183, train-acc 0.837, loss 0.415, para-norm 4952.668, grad-norm 3.412, time 6.95s, 
epoch 90, batches 12000|18183, train-acc 0.838, loss 0.417, para-norm 4952.704, grad-norm 37.591, time 6.95s, 
epoch 90, batches 13000|18183, train-acc 0.837, loss 0.418, para-norm 4952.350, grad-norm 3.706, time 6.93s, 
epoch 90, batches 14000|18183, train-acc 0.841, loss 0.413, para-norm 4952.288, grad-norm 3.786, time 6.95s, 
epoch 90, batches 15000|18183, train-acc 0.842, loss 0.411, para-norm 4951.771, grad-norm 1.872, time 6.95s, 
epoch 90, batches 16000|18183, train-acc 0.836, loss 0.421, para-norm 4950.724, grad-norm 4.158, time 6.95s, 
epoch 90, batches 17000|18183, train-acc 0.836, loss 0.423, para-norm 4950.443, grad-norm 2.362, time 6.95s, 
epoch 90, batches 18000|18183, train-acc 0.841, loss 0.413, para-norm 4950.694, grad-norm 3.918, time 6.95s, 
epoch 90, batches 18183|18183, train-acc 0.847, loss 0.400, para-norm 4950.779, grad-norm 0.606, time 1.27s, 
dev-acc 0.851
epoch 91, batches 1000|18183, train-acc 0.840, loss 0.415, para-norm 4950.289, grad-norm 4.714, time 7.07s, 
epoch 91, batches 2000|18183, train-acc 0.841, loss 0.413, para-norm 4958.277, grad-norm 3.579, time 6.96s, 
epoch 91, batches 3000|18183, train-acc 0.837, loss 0.421, para-norm 4959.980, grad-norm 4.300, time 6.96s, 
epoch 91, batches 4000|18183, train-acc 0.839, loss 0.417, para-norm 4960.639, grad-norm 4.158, time 6.95s, 
epoch 91, batches 5000|18183, train-acc 0.838, loss 0.418, para-norm 4959.996, grad-norm 3.621, time 6.94s, 
epoch 91, batches 6000|18183, train-acc 0.840, loss 0.418, para-norm 4960.079, grad-norm 4.154, time 6.95s, 
epoch 91, batches 7000|18183, train-acc 0.838, loss 0.416, para-norm 4967.753, grad-norm 6.381, time 6.94s, 
epoch 91, batches 8000|18183, train-acc 0.835, loss 0.421, para-norm 4968.244, grad-norm 2.633, time 7.01s, 
epoch 91, batches 9000|18183, train-acc 0.844, loss 0.404, para-norm 4966.926, grad-norm 1.358, time 6.94s, 
epoch 91, batches 10000|18183, train-acc 0.837, loss 0.416, para-norm 4966.167, grad-norm 2.238, time 6.94s, 
epoch 91, batches 11000|18183, train-acc 0.839, loss 0.415, para-norm 4965.957, grad-norm 2.311, time 6.94s, 
epoch 91, batches 12000|18183, train-acc 0.837, loss 0.420, para-norm 4967.000, grad-norm 5.352, time 6.95s, 
epoch 91, batches 13000|18183, train-acc 0.844, loss 0.409, para-norm 4966.303, grad-norm 1.704, time 6.94s, 
epoch 91, batches 14000|18183, train-acc 0.839, loss 0.421, para-norm 4965.461, grad-norm 4.659, time 6.94s, 
epoch 91, batches 15000|18183, train-acc 0.841, loss 0.415, para-norm 4965.513, grad-norm 2.540, time 6.95s, 
epoch 91, batches 16000|18183, train-acc 0.842, loss 0.408, para-norm 4964.965, grad-norm 3.885, time 6.93s, 
epoch 91, batches 17000|18183, train-acc 0.839, loss 0.418, para-norm 4965.055, grad-norm 5.538, time 6.94s, 
epoch 91, batches 18000|18183, train-acc 0.840, loss 0.413, para-norm 4964.048, grad-norm 5.995, time 6.96s, 
epoch 91, batches 18183|18183, train-acc 0.840, loss 0.410, para-norm 4964.240, grad-norm 4.617, time 1.27s, 
dev-acc 0.850
epoch 92, batches 1000|18183, train-acc 0.842, loss 0.408, para-norm 4964.360, grad-norm 1.159, time 7.06s, 
epoch 92, batches 2000|18183, train-acc 0.840, loss 0.413, para-norm 4973.018, grad-norm 4.433, time 6.96s, 
epoch 92, batches 3000|18183, train-acc 0.838, loss 0.418, para-norm 4980.575, grad-norm 3.563, time 6.95s, 
epoch 92, batches 4000|18183, train-acc 0.841, loss 0.414, para-norm 4980.835, grad-norm 4.460, time 6.95s, 
epoch 92, batches 5000|18183, train-acc 0.838, loss 0.415, para-norm 4981.263, grad-norm 2.396, time 6.94s, 
epoch 92, batches 6000|18183, train-acc 0.836, loss 0.422, para-norm 4979.503, grad-norm 5.238, time 6.94s, 
epoch 92, batches 7000|18183, train-acc 0.844, loss 0.411, para-norm 4979.654, grad-norm 4.735, time 6.93s, 
epoch 92, batches 8000|18183, train-acc 0.840, loss 0.413, para-norm 4980.139, grad-norm 3.914, time 6.95s, 
epoch 92, batches 9000|18183, train-acc 0.837, loss 0.420, para-norm 4979.817, grad-norm 4.688, time 6.95s, 
epoch 92, batches 10000|18183, train-acc 0.839, loss 0.415, para-norm 4986.007, grad-norm 4.604, time 6.94s, 
epoch 92, batches 11000|18183, train-acc 0.840, loss 0.411, para-norm 4987.154, grad-norm 354.265, time 6.96s, 
epoch 92, batches 12000|18183, train-acc 0.841, loss 0.409, para-norm 4984.821, grad-norm 1.973, time 6.94s, 
epoch 92, batches 13000|18183, train-acc 0.838, loss 0.419, para-norm 4985.515, grad-norm 7.402, time 6.95s, 
epoch 92, batches 14000|18183, train-acc 0.838, loss 0.416, para-norm 4986.011, grad-norm 3.289, time 6.95s, 
epoch 92, batches 15000|18183, train-acc 0.841, loss 0.410, para-norm 4985.401, grad-norm 3.199, time 6.95s, 
epoch 92, batches 16000|18183, train-acc 0.840, loss 0.412, para-norm 4983.966, grad-norm 6.105, time 6.94s, 
epoch 92, batches 17000|18183, train-acc 0.837, loss 0.421, para-norm 4982.305, grad-norm 5.952, time 6.94s, 
epoch 92, batches 18000|18183, train-acc 0.838, loss 0.417, para-norm 4981.892, grad-norm 3.331, time 6.95s, 
epoch 92, batches 18183|18183, train-acc 0.847, loss 0.401, para-norm 4982.505, grad-norm 4.530, time 1.27s, 
dev-acc 0.850
epoch 93, batches 1000|18183, train-acc 0.840, loss 0.415, para-norm 4982.431, grad-norm 7.524, time 7.04s, 
epoch 93, batches 2000|18183, train-acc 0.837, loss 0.415, para-norm 4982.153, grad-norm 2.004, time 6.94s, 
epoch 93, batches 3000|18183, train-acc 0.839, loss 0.412, para-norm 4981.256, grad-norm 5.460, time 6.95s, 
epoch 93, batches 4000|18183, train-acc 0.841, loss 0.410, para-norm 4980.075, grad-norm 868.340, time 6.94s, 
epoch 93, batches 5000|18183, train-acc 0.845, loss 0.407, para-norm 4980.184, grad-norm 2.776, time 6.94s, 
epoch 93, batches 6000|18183, train-acc 0.840, loss 0.412, para-norm 4980.683, grad-norm 4.770, time 6.95s, 
epoch 93, batches 7000|18183, train-acc 0.837, loss 0.416, para-norm 4980.740, grad-norm 3.079, time 6.94s, 
epoch 93, batches 8000|18183, train-acc 0.840, loss 0.415, para-norm 4981.161, grad-norm 5.076, time 6.94s, 
epoch 93, batches 9000|18183, train-acc 0.843, loss 0.410, para-norm 4981.434, grad-norm 2.961, time 6.96s, 
epoch 93, batches 10000|18183, train-acc 0.842, loss 0.409, para-norm 4981.316, grad-norm 5.238, time 6.94s, 
epoch 93, batches 11000|18183, train-acc 0.839, loss 0.413, para-norm 4980.852, grad-norm 5.667, time 6.95s, 
epoch 93, batches 12000|18183, train-acc 0.843, loss 0.412, para-norm 4981.017, grad-norm 5.453, time 6.94s, 
epoch 93, batches 13000|18183, train-acc 0.838, loss 0.416, para-norm 4980.348, grad-norm 3.875, time 6.95s, 
epoch 93, batches 14000|18183, train-acc 0.840, loss 0.413, para-norm 4980.470, grad-norm 1.309, time 6.96s, 
epoch 93, batches 15000|18183, train-acc 0.839, loss 0.413, para-norm 4979.897, grad-norm 2.814, time 6.96s, 
epoch 93, batches 16000|18183, train-acc 0.842, loss 0.411, para-norm 4979.150, grad-norm 2.972, time 6.94s, 
epoch 93, batches 17000|18183, train-acc 0.839, loss 0.414, para-norm 4978.464, grad-norm 1.382, time 6.95s, 
epoch 93, batches 18000|18183, train-acc 0.835, loss 0.423, para-norm 4977.821, grad-norm 4.404, time 6.96s, 
epoch 93, batches 18183|18183, train-acc 0.841, loss 0.414, para-norm 4978.097, grad-norm 2.461, time 1.27s, 
dev-acc 0.854
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
save model!
epoch 94, batches 1000|18183, train-acc 0.846, loss 0.404, para-norm 4977.987, grad-norm 2.856, time 7.24s, 
epoch 94, batches 2000|18183, train-acc 0.842, loss 0.408, para-norm 4976.921, grad-norm 1.393, time 7.21s, 
epoch 94, batches 3000|18183, train-acc 0.844, loss 0.403, para-norm 4984.671, grad-norm 2.512, time 7.21s, 
epoch 94, batches 4000|18183, train-acc 0.840, loss 0.416, para-norm 4985.604, grad-norm 3.198, time 7.21s, 
epoch 94, batches 5000|18183, train-acc 0.840, loss 0.415, para-norm 4985.504, grad-norm 3.193, time 7.20s, 
epoch 94, batches 6000|18183, train-acc 0.841, loss 0.410, para-norm 4984.992, grad-norm 3.528, time 7.20s, 
epoch 94, batches 7000|18183, train-acc 0.844, loss 0.407, para-norm 4984.387, grad-norm 6.590, time 7.20s, 
epoch 94, batches 8000|18183, train-acc 0.839, loss 0.415, para-norm 4984.319, grad-norm 3.655, time 7.20s, 
epoch 94, batches 9000|18183, train-acc 0.840, loss 0.414, para-norm 4984.727, grad-norm 1.951, time 7.20s, 
epoch 94, batches 10000|18183, train-acc 0.838, loss 0.418, para-norm 4984.427, grad-norm 4.186, time 7.20s, 
epoch 94, batches 11000|18183, train-acc 0.838, loss 0.410, para-norm 4984.075, grad-norm 4.178, time 7.19s, 
epoch 94, batches 12000|18183, train-acc 0.841, loss 0.412, para-norm 4984.252, grad-norm 4.898, time 7.19s, 
epoch 94, batches 13000|18183, train-acc 0.841, loss 0.414, para-norm 4983.757, grad-norm 2.004, time 7.21s, 
epoch 94, batches 14000|18183, train-acc 0.839, loss 0.412, para-norm 4983.150, grad-norm 4.751, time 7.20s, 
epoch 94, batches 15000|18183, train-acc 0.835, loss 0.421, para-norm 4982.037, grad-norm 0.727, time 7.20s, 
epoch 94, batches 16000|18183, train-acc 0.839, loss 0.417, para-norm 4981.522, grad-norm 2.741, time 7.20s, 
epoch 94, batches 17000|18183, train-acc 0.839, loss 0.413, para-norm 4980.273, grad-norm 2.454, time 7.19s, 
epoch 94, batches 18000|18183, train-acc 0.839, loss 0.413, para-norm 4980.605, grad-norm 3.633, time 7.21s, 
epoch 94, batches 18183|18183, train-acc 0.836, loss 0.423, para-norm 4980.783, grad-norm 3.595, time 1.31s, 
dev-acc 0.854
epoch 95, batches 1000|18183, train-acc 0.842, loss 0.407, para-norm 4980.549, grad-norm 3.218, time 7.27s, 
epoch 95, batches 2000|18183, train-acc 0.841, loss 0.409, para-norm 4980.690, grad-norm 1.456, time 7.44s, 
epoch 95, batches 3000|18183, train-acc 0.842, loss 0.406, para-norm 4981.052, grad-norm 2.828, time 7.75s, 
epoch 95, batches 4000|18183, train-acc 0.845, loss 0.405, para-norm 4979.814, grad-norm 6.638, time 7.82s, 
epoch 95, batches 5000|18183, train-acc 0.838, loss 0.415, para-norm 4980.689, grad-norm 5.984, time 7.76s, 
epoch 95, batches 6000|18183, train-acc 0.841, loss 0.411, para-norm 4980.791, grad-norm 5.458, time 7.76s, 
epoch 95, batches 7000|18183, train-acc 0.841, loss 0.410, para-norm 4981.812, grad-norm 1.657, time 7.79s, 
epoch 95, batches 8000|18183, train-acc 0.842, loss 0.409, para-norm 4981.570, grad-norm 4.384, time 7.77s, 
epoch 95, batches 9000|18183, train-acc 0.839, loss 0.416, para-norm 4987.092, grad-norm 5.645, time 7.78s, 
epoch 95, batches 10000|18183, train-acc 0.843, loss 0.405, para-norm 4987.122, grad-norm 2.048, time 7.77s, 
epoch 95, batches 11000|18183, train-acc 0.843, loss 0.409, para-norm 4987.763, grad-norm 3.112, time 7.77s, 
epoch 95, batches 12000|18183, train-acc 0.839, loss 0.418, para-norm 4986.723, grad-norm 4.244, time 7.78s, 
epoch 95, batches 13000|18183, train-acc 0.839, loss 0.417, para-norm 4986.562, grad-norm 3.357, time 7.78s, 
epoch 95, batches 14000|18183, train-acc 0.843, loss 0.406, para-norm 4987.257, grad-norm 5.597, time 7.76s, 
epoch 95, batches 15000|18183, train-acc 0.838, loss 0.415, para-norm 4987.446, grad-norm 3.991, time 7.78s, 
epoch 95, batches 16000|18183, train-acc 0.842, loss 0.406, para-norm 4986.553, grad-norm 1.593, time 7.34s, 
epoch 95, batches 17000|18183, train-acc 0.840, loss 0.414, para-norm 4986.202, grad-norm 2.882, time 7.28s, 
epoch 95, batches 18000|18183, train-acc 0.839, loss 0.411, para-norm 4985.956, grad-norm 67.945, time 7.21s, 
epoch 95, batches 18183|18183, train-acc 0.840, loss 0.411, para-norm 4985.521, grad-norm 3.800, time 1.32s, 
dev-acc 0.852
epoch 96, batches 1000|18183, train-acc 0.843, loss 0.404, para-norm 4985.885, grad-norm 5.820, time 7.28s, 
epoch 96, batches 2000|18183, train-acc 0.842, loss 0.409, para-norm 4986.144, grad-norm 3.223, time 7.21s, 
epoch 96, batches 3000|18183, train-acc 0.842, loss 0.410, para-norm 4986.764, grad-norm 32.165, time 7.21s, 
epoch 96, batches 4000|18183, train-acc 0.841, loss 0.411, para-norm 4986.780, grad-norm 5.088, time 7.21s, 
epoch 96, batches 5000|18183, train-acc 0.842, loss 0.412, para-norm 4986.090, grad-norm 1.904, time 7.20s, 
epoch 96, batches 6000|18183, train-acc 0.843, loss 0.402, para-norm 4985.454, grad-norm 2.717, time 7.21s, 
epoch 96, batches 7000|18183, train-acc 0.842, loss 0.409, para-norm 4984.622, grad-norm 4.429, time 7.21s, 
epoch 96, batches 8000|18183, train-acc 0.839, loss 0.416, para-norm 4984.104, grad-norm 2.799, time 7.20s, 
epoch 96, batches 9000|18183, train-acc 0.844, loss 0.404, para-norm 4983.998, grad-norm 2.441, time 7.22s, 
epoch 96, batches 10000|18183, train-acc 0.842, loss 0.408, para-norm 4983.744, grad-norm 2.713, time 7.20s, 
epoch 96, batches 11000|18183, train-acc 0.841, loss 0.413, para-norm 4983.877, grad-norm 1.915, time 7.21s, 
epoch 96, batches 12000|18183, train-acc 0.841, loss 0.411, para-norm 4983.830, grad-norm 6.508, time 7.21s, 
epoch 96, batches 13000|18183, train-acc 0.842, loss 0.408, para-norm 4983.520, grad-norm 2.262, time 7.21s, 
epoch 96, batches 14000|18183, train-acc 0.843, loss 0.407, para-norm 4984.034, grad-norm 2.005, time 7.21s, 
epoch 96, batches 15000|18183, train-acc 0.844, loss 0.406, para-norm 4982.945, grad-norm 3.784, time 7.21s, 
epoch 96, batches 16000|18183, train-acc 0.841, loss 0.415, para-norm 4982.361, grad-norm 3.514, time 7.21s, 
epoch 96, batches 17000|18183, train-acc 0.840, loss 0.418, para-norm 4981.998, grad-norm 5.315, time 7.21s, 
epoch 96, batches 18000|18183, train-acc 0.842, loss 0.410, para-norm 4982.424, grad-norm 2.167, time 7.28s, 
epoch 96, batches 18183|18183, train-acc 0.835, loss 0.418, para-norm 4981.987, grad-norm 4.268, time 1.32s, 
dev-acc 0.857
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
save model!
epoch 97, batches 1000|18183, train-acc 0.845, loss 0.401, para-norm 4982.888, grad-norm 8.245, time 7.36s, 
epoch 97, batches 2000|18183, train-acc 0.844, loss 0.405, para-norm 4983.021, grad-norm 1.439, time 7.23s, 
epoch 97, batches 3000|18183, train-acc 0.844, loss 0.405, para-norm 4982.899, grad-norm 5.819, time 7.22s, 
epoch 97, batches 4000|18183, train-acc 0.841, loss 0.409, para-norm 4982.038, grad-norm 3.975, time 7.22s, 
epoch 97, batches 5000|18183, train-acc 0.845, loss 0.403, para-norm 4982.527, grad-norm 3.848, time 7.21s, 
epoch 97, batches 6000|18183, train-acc 0.838, loss 0.414, para-norm 4982.574, grad-norm 3.546, time 7.22s, 
epoch 97, batches 7000|18183, train-acc 0.844, loss 0.404, para-norm 4982.957, grad-norm 1.968, time 7.21s, 
epoch 97, batches 8000|18183, train-acc 0.843, loss 0.403, para-norm 4982.757, grad-norm 4.110, time 7.22s, 
epoch 97, batches 9000|18183, train-acc 0.838, loss 0.412, para-norm 4990.005, grad-norm 2.780, time 7.23s, 
epoch 97, batches 10000|18183, train-acc 0.844, loss 0.403, para-norm 4991.747, grad-norm 2.451, time 7.21s, 
epoch 97, batches 11000|18183, train-acc 0.839, loss 0.411, para-norm 4991.607, grad-norm 1.639, time 7.22s, 
epoch 97, batches 12000|18183, train-acc 0.841, loss 0.410, para-norm 4992.090, grad-norm 3.684, time 7.21s, 
epoch 97, batches 13000|18183, train-acc 0.840, loss 0.413, para-norm 4991.543, grad-norm 11.512, time 7.22s, 
epoch 97, batches 14000|18183, train-acc 0.839, loss 0.415, para-norm 4990.795, grad-norm 5.197, time 7.21s, 
epoch 97, batches 15000|18183, train-acc 0.842, loss 0.405, para-norm 4991.010, grad-norm 3.054, time 7.20s, 
epoch 97, batches 16000|18183, train-acc 0.837, loss 0.417, para-norm 4991.073, grad-norm 8.174, time 7.21s, 
epoch 97, batches 17000|18183, train-acc 0.840, loss 0.415, para-norm 4990.699, grad-norm 1.929, time 7.21s, 
epoch 97, batches 18000|18183, train-acc 0.839, loss 0.415, para-norm 4990.062, grad-norm 6.232, time 7.20s, 
epoch 97, batches 18183|18183, train-acc 0.842, loss 0.413, para-norm 4990.509, grad-norm 4.002, time 1.31s, 
dev-acc 0.853
epoch 98, batches 1000|18183, train-acc 0.842, loss 0.407, para-norm 4990.873, grad-norm 2.484, time 7.29s, 
epoch 98, batches 2000|18183, train-acc 0.844, loss 0.401, para-norm 4999.957, grad-norm 2.974, time 7.29s, 
epoch 98, batches 3000|18183, train-acc 0.844, loss 0.406, para-norm 4999.169, grad-norm 5.843, time 7.21s, 
epoch 98, batches 4000|18183, train-acc 0.845, loss 0.399, para-norm 4999.412, grad-norm 4.122, time 7.21s, 
epoch 98, batches 5000|18183, train-acc 0.842, loss 0.405, para-norm 4999.012, grad-norm 5.428, time 7.20s, 
epoch 98, batches 6000|18183, train-acc 0.841, loss 0.408, para-norm 4998.524, grad-norm 4.658, time 7.22s, 
epoch 98, batches 7000|18183, train-acc 0.839, loss 0.415, para-norm 4998.380, grad-norm 6.181, time 7.22s, 
epoch 98, batches 8000|18183, train-acc 0.839, loss 0.411, para-norm 4997.211, grad-norm 2.747, time 7.21s, 
epoch 98, batches 9000|18183, train-acc 0.841, loss 0.413, para-norm 4997.204, grad-norm 1.426, time 7.20s, 
epoch 98, batches 10000|18183, train-acc 0.847, loss 0.396, para-norm 4997.215, grad-norm 2.695, time 7.21s, 
epoch 98, batches 11000|18183, train-acc 0.845, loss 0.405, para-norm 4997.028, grad-norm 4.931, time 7.22s, 
epoch 98, batches 12000|18183, train-acc 0.838, loss 0.414, para-norm 4996.440, grad-norm 7.150, time 7.21s, 
epoch 98, batches 13000|18183, train-acc 0.843, loss 0.407, para-norm 4995.795, grad-norm 3.199, time 7.20s, 
epoch 98, batches 14000|18183, train-acc 0.844, loss 0.404, para-norm 4994.700, grad-norm 4.230, time 7.22s, 
epoch 98, batches 15000|18183, train-acc 0.842, loss 0.409, para-norm 4994.714, grad-norm 2.259, time 7.21s, 
epoch 98, batches 16000|18183, train-acc 0.844, loss 0.405, para-norm 4995.131, grad-norm 123.337, time 7.21s, 
epoch 98, batches 17000|18183, train-acc 0.841, loss 0.415, para-norm 4994.377, grad-norm 3.267, time 7.22s, 
epoch 98, batches 18000|18183, train-acc 0.842, loss 0.410, para-norm 5001.765, grad-norm 2.082, time 7.21s, 
epoch 98, batches 18183|18183, train-acc 0.839, loss 0.423, para-norm 5001.775, grad-norm 3.028, time 1.31s, 
dev-acc 0.855
epoch 99, batches 1000|18183, train-acc 0.841, loss 0.410, para-norm 5002.837, grad-norm 3.234, time 7.29s, 
epoch 99, batches 2000|18183, train-acc 0.841, loss 0.409, para-norm 5011.128, grad-norm 2.825, time 7.20s, 
epoch 99, batches 3000|18183, train-acc 0.843, loss 0.410, para-norm 5011.937, grad-norm 3.644, time 7.20s, 
epoch 99, batches 4000|18183, train-acc 0.841, loss 0.403, para-norm 5011.047, grad-norm 4.009, time 7.22s, 
epoch 99, batches 5000|18183, train-acc 0.850, loss 0.395, para-norm 5010.089, grad-norm 1.231, time 7.21s, 
epoch 99, batches 6000|18183, train-acc 0.844, loss 0.408, para-norm 5009.905, grad-norm 3.146, time 7.20s, 
epoch 99, batches 7000|18183, train-acc 0.844, loss 0.403, para-norm 5010.091, grad-norm 2.384, time 7.21s, 
epoch 99, batches 8000|18183, train-acc 0.841, loss 0.408, para-norm 5010.843, grad-norm 39.869, time 7.21s, 
epoch 99, batches 9000|18183, train-acc 0.843, loss 0.406, para-norm 5013.466, grad-norm 3387.084, time 7.22s, 
epoch 99, batches 10000|18183, train-acc 0.839, loss 0.411, para-norm 5013.639, grad-norm 3.272, time 7.22s, 
epoch 99, batches 11000|18183, train-acc 0.842, loss 0.406, para-norm 5012.754, grad-norm 8.231, time 7.21s, 
epoch 99, batches 12000|18183, train-acc 0.843, loss 0.406, para-norm 5011.644, grad-norm 3.416, time 7.21s, 
epoch 99, batches 13000|18183, train-acc 0.841, loss 0.412, para-norm 5011.205, grad-norm 2.620, time 7.21s, 
epoch 99, batches 14000|18183, train-acc 0.844, loss 0.404, para-norm 5010.547, grad-norm 4.669, time 7.21s, 
epoch 99, batches 15000|18183, train-acc 0.843, loss 0.410, para-norm 5009.661, grad-norm 12.669, time 7.21s, 
epoch 99, batches 16000|18183, train-acc 0.844, loss 0.405, para-norm 5009.778, grad-norm 1.466, time 7.22s, 
epoch 99, batches 17000|18183, train-acc 0.841, loss 0.408, para-norm 5009.524, grad-norm 30.225, time 7.23s, 
epoch 99, batches 18000|18183, train-acc 0.842, loss 0.412, para-norm 5009.152, grad-norm 1.726, time 7.28s, 
epoch 99, batches 18183|18183, train-acc 0.845, loss 0.403, para-norm 5009.323, grad-norm 2.690, time 1.32s, 
dev-acc 0.853
epoch 100, batches 1000|18183, train-acc 0.845, loss 0.405, para-norm 5009.843, grad-norm 4.184, time 7.35s, 
epoch 100, batches 2000|18183, train-acc 0.841, loss 0.408, para-norm 5009.895, grad-norm 3.080, time 7.20s, 
epoch 100, batches 3000|18183, train-acc 0.845, loss 0.403, para-norm 5009.697, grad-norm 1.164, time 7.21s, 
epoch 100, batches 4000|18183, train-acc 0.842, loss 0.408, para-norm 5008.710, grad-norm 3.820, time 7.21s, 
epoch 100, batches 5000|18183, train-acc 0.843, loss 0.408, para-norm 5009.304, grad-norm 0.983, time 7.22s, 
epoch 100, batches 6000|18183, train-acc 0.846, loss 0.402, para-norm 5008.062, grad-norm 2.887, time 7.20s, 
epoch 100, batches 7000|18183, train-acc 0.839, loss 0.411, para-norm 5008.729, grad-norm 5.091, time 7.21s, 
epoch 100, batches 8000|18183, train-acc 0.846, loss 0.406, para-norm 5008.057, grad-norm 4.531, time 7.22s, 
epoch 100, batches 9000|18183, train-acc 0.845, loss 0.401, para-norm 5007.791, grad-norm 3.717, time 7.21s, 
epoch 100, batches 10000|18183, train-acc 0.846, loss 0.402, para-norm 5008.205, grad-norm 5.556, time 7.22s, 
epoch 100, batches 11000|18183, train-acc 0.843, loss 0.405, para-norm 5007.206, grad-norm 1.325, time 7.22s, 
epoch 100, batches 12000|18183, train-acc 0.841, loss 0.407, para-norm 5007.166, grad-norm 4.254, time 7.22s, 
epoch 100, batches 13000|18183, train-acc 0.846, loss 0.399, para-norm 5007.453, grad-norm 3.882, time 7.21s, 
epoch 100, batches 14000|18183, train-acc 0.842, loss 0.406, para-norm 5006.282, grad-norm 1.764, time 7.22s, 
epoch 100, batches 15000|18183, train-acc 0.840, loss 0.410, para-norm 5006.702, grad-norm 8.816, time 7.22s, 
epoch 100, batches 16000|18183, train-acc 0.843, loss 0.405, para-norm 5006.516, grad-norm 1.507, time 7.21s, 
epoch 100, batches 17000|18183, train-acc 0.845, loss 0.403, para-norm 5006.050, grad-norm 4.989, time 7.21s, 
epoch 100, batches 18000|18183, train-acc 0.843, loss 0.406, para-norm 5005.391, grad-norm 1.706, time 7.20s, 
epoch 100, batches 18183|18183, train-acc 0.833, loss 0.423, para-norm 5004.862, grad-norm 5.358, time 1.32s, 
dev-acc 0.855
epoch 101, batches 1000|18183, train-acc 0.846, loss 0.395, para-norm 5004.962, grad-norm 3.626, time 7.28s, 
epoch 101, batches 2000|18183, train-acc 0.845, loss 0.399, para-norm 5005.794, grad-norm 3.223, time 7.20s, 
epoch 101, batches 3000|18183, train-acc 0.844, loss 0.406, para-norm 5005.483, grad-norm 4.235, time 7.21s, 
epoch 101, batches 4000|18183, train-acc 0.839, loss 0.411, para-norm 5006.373, grad-norm 1.257, time 7.22s, 
epoch 101, batches 5000|18183, train-acc 0.845, loss 0.401, para-norm 5006.378, grad-norm 1.333, time 7.19s, 
epoch 101, batches 6000|18183, train-acc 0.842, loss 0.411, para-norm 5006.232, grad-norm 4.534, time 7.21s, 
epoch 101, batches 7000|18183, train-acc 0.846, loss 0.402, para-norm 5006.758, grad-norm 0.927, time 7.20s, 
epoch 101, batches 8000|18183, train-acc 0.844, loss 0.404, para-norm 5005.310, grad-norm 2.987, time 7.22s, 
epoch 101, batches 9000|18183, train-acc 0.841, loss 0.410, para-norm 5005.121, grad-norm 1.979, time 7.22s, 
epoch 101, batches 10000|18183, train-acc 0.843, loss 0.407, para-norm 5004.822, grad-norm 5.809, time 7.21s, 
epoch 101, batches 11000|18183, train-acc 0.843, loss 0.409, para-norm 5004.454, grad-norm 2.530, time 7.21s, 
epoch 101, batches 12000|18183, train-acc 0.845, loss 0.402, para-norm 5004.340, grad-norm 1.323, time 7.20s, 
epoch 101, batches 13000|18183, train-acc 0.840, loss 0.411, para-norm 5004.083, grad-norm 5.579, time 7.21s, 
epoch 101, batches 14000|18183, train-acc 0.843, loss 0.405, para-norm 5004.428, grad-norm 2.318, time 7.21s, 
epoch 101, batches 15000|18183, train-acc 0.844, loss 0.406, para-norm 5004.760, grad-norm 3.994, time 7.21s, 
epoch 101, batches 16000|18183, train-acc 0.841, loss 0.408, para-norm 5004.054, grad-norm 5.681, time 7.22s, 
epoch 101, batches 17000|18183, train-acc 0.842, loss 0.413, para-norm 5004.849, grad-norm 4.224, time 7.21s, 
epoch 101, batches 18000|18183, train-acc 0.844, loss 0.409, para-norm 5004.155, grad-norm 4.623, time 7.21s, 
epoch 101, batches 18183|18183, train-acc 0.848, loss 0.392, para-norm 5004.101, grad-norm 2.638, time 1.32s, 
dev-acc 0.852
epoch 102, batches 1000|18183, train-acc 0.844, loss 0.400, para-norm 5004.679, grad-norm 4.438, time 7.27s, 
epoch 102, batches 2000|18183, train-acc 0.842, loss 0.407, para-norm 5004.929, grad-norm 2.514, time 7.21s, 
epoch 102, batches 3000|18183, train-acc 0.848, loss 0.397, para-norm 5005.948, grad-norm 3.815, time 7.21s, 
epoch 102, batches 4000|18183, train-acc 0.839, loss 0.410, para-norm 5013.879, grad-norm 1.722, time 7.22s, 
epoch 102, batches 5000|18183, train-acc 0.840, loss 0.410, para-norm 5012.903, grad-norm 6.472, time 7.22s, 
epoch 102, batches 6000|18183, train-acc 0.841, loss 0.413, para-norm 5011.996, grad-norm 2.584, time 7.22s, 
epoch 102, batches 7000|18183, train-acc 0.849, loss 0.398, para-norm 5014.247, grad-norm 2.661, time 7.21s, 
epoch 102, batches 8000|18183, train-acc 0.841, loss 0.409, para-norm 5014.304, grad-norm 1.764, time 7.21s, 
epoch 102, batches 9000|18183, train-acc 0.842, loss 0.407, para-norm 5013.366, grad-norm 3.457, time 7.22s, 
epoch 102, batches 10000|18183, train-acc 0.848, loss 0.398, para-norm 5012.852, grad-norm 4.333, time 7.21s, 
epoch 102, batches 11000|18183, train-acc 0.845, loss 0.408, para-norm 5012.409, grad-norm 2.488, time 7.22s, 
epoch 102, batches 12000|18183, train-acc 0.844, loss 0.403, para-norm 5011.987, grad-norm 5.921, time 7.22s, 
epoch 102, batches 13000|18183, train-acc 0.843, loss 0.409, para-norm 5011.291, grad-norm 5.659, time 7.22s, 
epoch 102, batches 14000|18183, train-acc 0.842, loss 0.407, para-norm 5010.155, grad-norm 96.416, time 7.21s, 
epoch 102, batches 15000|18183, train-acc 0.844, loss 0.401, para-norm 5009.873, grad-norm 5.619, time 7.20s, 
epoch 102, batches 16000|18183, train-acc 0.844, loss 0.405, para-norm 5010.498, grad-norm 3.772, time 7.21s, 
epoch 102, batches 17000|18183, train-acc 0.847, loss 0.400, para-norm 5010.298, grad-norm 2.346, time 7.21s, 
epoch 102, batches 18000|18183, train-acc 0.845, loss 0.405, para-norm 5009.854, grad-norm 3.642, time 7.21s, 
epoch 102, batches 18183|18183, train-acc 0.851, loss 0.392, para-norm 5010.246, grad-norm 10.956, time 1.31s, 
dev-acc 0.853
epoch 103, batches 1000|18183, train-acc 0.844, loss 0.405, para-norm 5010.745, grad-norm 4.643, time 7.26s, 
epoch 103, batches 2000|18183, train-acc 0.846, loss 0.399, para-norm 5011.063, grad-norm 3.935, time 7.21s, 
epoch 103, batches 3000|18183, train-acc 0.845, loss 0.400, para-norm 5011.849, grad-norm 1.551, time 7.21s, 
epoch 103, batches 4000|18183, train-acc 0.846, loss 0.397, para-norm 5012.601, grad-norm 6.116, time 7.21s, 
epoch 103, batches 5000|18183, train-acc 0.844, loss 0.405, para-norm 5011.881, grad-norm 4.084, time 7.21s, 
epoch 103, batches 6000|18183, train-acc 0.846, loss 0.398, para-norm 5011.458, grad-norm 6.594, time 7.22s, 
epoch 103, batches 7000|18183, train-acc 0.844, loss 0.403, para-norm 5016.623, grad-norm 6.128, time 7.22s, 
epoch 103, batches 8000|18183, train-acc 0.844, loss 0.408, para-norm 5016.521, grad-norm 5.664, time 7.21s, 
epoch 103, batches 9000|18183, train-acc 0.845, loss 0.404, para-norm 5016.157, grad-norm 2.790, time 7.21s, 
epoch 103, batches 10000|18183, train-acc 0.839, loss 0.411, para-norm 5016.498, grad-norm 2.974, time 7.22s, 
epoch 103, batches 11000|18183, train-acc 0.844, loss 0.405, para-norm 5017.113, grad-norm 1.994, time 7.21s, 
epoch 103, batches 12000|18183, train-acc 0.846, loss 0.403, para-norm 5016.309, grad-norm 3.440, time 7.21s, 
epoch 103, batches 13000|18183, train-acc 0.843, loss 0.404, para-norm 5015.731, grad-norm 2.416, time 7.20s, 
epoch 103, batches 14000|18183, train-acc 0.843, loss 0.406, para-norm 5016.680, grad-norm 1.485, time 7.27s, 
epoch 103, batches 15000|18183, train-acc 0.844, loss 0.405, para-norm 5016.416, grad-norm 6.456, time 7.22s, 
epoch 103, batches 16000|18183, train-acc 0.846, loss 0.402, para-norm 5015.262, grad-norm 4.842, time 7.21s, 
epoch 103, batches 17000|18183, train-acc 0.845, loss 0.407, para-norm 5014.916, grad-norm 1.857, time 7.22s, 
epoch 103, batches 18000|18183, train-acc 0.844, loss 0.403, para-norm 5015.049, grad-norm 5.428, time 7.22s, 
epoch 103, batches 18183|18183, train-acc 0.841, loss 0.407, para-norm 5014.686, grad-norm 2.707, time 1.32s, 
dev-acc 0.855
epoch 104, batches 1000|18183, train-acc 0.847, loss 0.399, para-norm 5015.136, grad-norm 2.112, time 7.28s, 
epoch 104, batches 2000|18183, train-acc 0.849, loss 0.393, para-norm 5014.947, grad-norm 1.079, time 7.20s, 
epoch 104, batches 3000|18183, train-acc 0.846, loss 0.400, para-norm 5014.397, grad-norm 3.732, time 7.22s, 
epoch 104, batches 4000|18183, train-acc 0.842, loss 0.406, para-norm 5015.003, grad-norm 3.599, time 7.22s, 
epoch 104, batches 5000|18183, train-acc 0.842, loss 0.407, para-norm 5014.790, grad-norm 1.079, time 7.21s, 
epoch 104, batches 6000|18183, train-acc 0.848, loss 0.396, para-norm 5015.215, grad-norm 2.417, time 7.23s, 
epoch 104, batches 7000|18183, train-acc 0.845, loss 0.399, para-norm 5015.066, grad-norm 1.654, time 7.22s, 
epoch 104, batches 8000|18183, train-acc 0.848, loss 0.399, para-norm 5014.406, grad-norm 5.333, time 7.22s, 
epoch 104, batches 9000|18183, train-acc 0.842, loss 0.408, para-norm 5014.427, grad-norm 4.980, time 7.22s, 
epoch 104, batches 10000|18183, train-acc 0.845, loss 0.405, para-norm 5014.719, grad-norm 3.600, time 7.22s, 
epoch 104, batches 11000|18183, train-acc 0.843, loss 0.404, para-norm 5014.133, grad-norm 4.095, time 7.22s, 
epoch 104, batches 12000|18183, train-acc 0.844, loss 0.404, para-norm 5014.350, grad-norm 4.157, time 7.20s, 
epoch 104, batches 13000|18183, train-acc 0.848, loss 0.399, para-norm 5020.213, grad-norm 3.790, time 7.20s, 
epoch 104, batches 14000|18183, train-acc 0.844, loss 0.404, para-norm 5020.290, grad-norm 2.909, time 7.19s, 
epoch 104, batches 15000|18183, train-acc 0.844, loss 0.406, para-norm 5019.388, grad-norm 1.171, time 7.20s, 
epoch 104, batches 16000|18183, train-acc 0.847, loss 0.402, para-norm 5020.105, grad-norm 3.248, time 7.20s, 
epoch 104, batches 17000|18183, train-acc 0.841, loss 0.413, para-norm 5019.605, grad-norm 2.880, time 7.22s, 
epoch 104, batches 18000|18183, train-acc 0.844, loss 0.404, para-norm 5026.695, grad-norm 3.633, time 7.21s, 
epoch 104, batches 18183|18183, train-acc 0.837, loss 0.415, para-norm 5026.450, grad-norm 64.877, time 1.32s, 
dev-acc 0.854
epoch 105, batches 1000|18183, train-acc 0.843, loss 0.400, para-norm 5030.142, grad-norm 8.046, time 7.27s, 
epoch 105, batches 2000|18183, train-acc 0.849, loss 0.393, para-norm 5031.187, grad-norm 4.511, time 7.21s, 
epoch 105, batches 3000|18183, train-acc 0.846, loss 0.395, para-norm 5030.464, grad-norm 3.771, time 7.22s, 
epoch 105, batches 4000|18183, train-acc 0.848, loss 0.398, para-norm 5029.592, grad-norm 1.783, time 7.22s, 
epoch 105, batches 5000|18183, train-acc 0.847, loss 0.392, para-norm 5029.687, grad-norm 4.748, time 7.22s, 
epoch 105, batches 6000|18183, train-acc 0.847, loss 0.403, para-norm 5028.971, grad-norm 4.360, time 7.22s, 
epoch 105, batches 7000|18183, train-acc 0.847, loss 0.398, para-norm 5028.396, grad-norm 1.221, time 7.21s, 
epoch 105, batches 8000|18183, train-acc 0.846, loss 0.398, para-norm 5027.958, grad-norm 1.912, time 7.21s, 
epoch 105, batches 9000|18183, train-acc 0.842, loss 0.409, para-norm 5027.482, grad-norm 3.865, time 7.23s, 
epoch 105, batches 10000|18183, train-acc 0.846, loss 0.404, para-norm 5027.063, grad-norm 4.898, time 7.23s, 
epoch 105, batches 11000|18183, train-acc 0.846, loss 0.402, para-norm 5026.784, grad-norm 5.255, time 7.22s, 
epoch 105, batches 12000|18183, train-acc 0.847, loss 0.399, para-norm 5026.515, grad-norm 7.155, time 7.22s, 
epoch 105, batches 13000|18183, train-acc 0.844, loss 0.402, para-norm 5026.126, grad-norm 4.071, time 7.21s, 
epoch 105, batches 14000|18183, train-acc 0.843, loss 0.411, para-norm 5025.427, grad-norm 4.271, time 7.22s, 
epoch 105, batches 15000|18183, train-acc 0.843, loss 0.406, para-norm 5024.699, grad-norm 4.615, time 7.23s, 
epoch 105, batches 16000|18183, train-acc 0.847, loss 0.397, para-norm 5024.746, grad-norm 7.348, time 7.22s, 
epoch 105, batches 17000|18183, train-acc 0.840, loss 0.411, para-norm 5024.485, grad-norm 8.431, time 7.22s, 
epoch 105, batches 18000|18183, train-acc 0.844, loss 0.403, para-norm 5024.556, grad-norm 4.179, time 7.22s, 
epoch 105, batches 18183|18183, train-acc 0.849, loss 0.388, para-norm 5024.557, grad-norm 1.035, time 1.32s, 
dev-acc 0.856
epoch 106, batches 1000|18183, train-acc 0.843, loss 0.402, para-norm 5024.546, grad-norm 2.817, time 7.29s, 
epoch 106, batches 2000|18183, train-acc 0.850, loss 0.393, para-norm 5024.352, grad-norm 3.583, time 7.23s, 
epoch 106, batches 3000|18183, train-acc 0.848, loss 0.399, para-norm 5023.530, grad-norm 7.637, time 7.21s, 
epoch 106, batches 4000|18183, train-acc 0.843, loss 0.400, para-norm 5024.130, grad-norm 3.510, time 7.22s, 
epoch 106, batches 5000|18183, train-acc 0.849, loss 0.398, para-norm 5024.019, grad-norm 1.684, time 7.22s, 
epoch 106, batches 6000|18183, train-acc 0.848, loss 0.399, para-norm 5023.847, grad-norm 2.059, time 7.22s, 
epoch 106, batches 7000|18183, train-acc 0.844, loss 0.403, para-norm 5024.252, grad-norm 2.838, time 7.22s, 
epoch 106, batches 8000|18183, train-acc 0.845, loss 0.405, para-norm 5024.133, grad-norm 4.055, time 7.23s, 
epoch 106, batches 9000|18183, train-acc 0.847, loss 0.398, para-norm 5024.566, grad-norm 3.204, time 7.20s, 
epoch 106, batches 10000|18183, train-acc 0.846, loss 0.400, para-norm 5025.170, grad-norm 5.230, time 7.22s, 
epoch 106, batches 11000|18183, train-acc 0.848, loss 0.397, para-norm 5025.150, grad-norm 5.442, time 7.22s, 
epoch 106, batches 12000|18183, train-acc 0.847, loss 0.396, para-norm 5025.232, grad-norm 3.613, time 7.22s, 
epoch 106, batches 13000|18183, train-acc 0.846, loss 0.400, para-norm 5024.639, grad-norm 2.226, time 7.22s, 
epoch 106, batches 14000|18183, train-acc 0.842, loss 0.406, para-norm 5024.211, grad-norm 22.687, time 7.23s, 
epoch 106, batches 15000|18183, train-acc 0.847, loss 0.397, para-norm 5024.107, grad-norm 2.042, time 7.22s, 
epoch 106, batches 16000|18183, train-acc 0.843, loss 0.402, para-norm 5023.914, grad-norm 9.579, time 7.22s, 
epoch 106, batches 17000|18183, train-acc 0.843, loss 0.406, para-norm 5023.634, grad-norm 8.431, time 7.23s, 
epoch 106, batches 18000|18183, train-acc 0.852, loss 0.394, para-norm 5022.121, grad-norm 3.530, time 7.22s, 
epoch 106, batches 18183|18183, train-acc 0.847, loss 0.409, para-norm 5022.287, grad-norm 4.449, time 1.33s, 
dev-acc 0.855
epoch 107, batches 1000|18183, train-acc 0.847, loss 0.400, para-norm 5022.923, grad-norm 7.262, time 7.26s, 
epoch 107, batches 2000|18183, train-acc 0.850, loss 0.394, para-norm 5023.519, grad-norm 4.638, time 7.23s, 
epoch 107, batches 3000|18183, train-acc 0.846, loss 0.404, para-norm 5023.203, grad-norm 3.699, time 7.22s, 
epoch 107, batches 4000|18183, train-acc 0.847, loss 0.396, para-norm 5023.917, grad-norm 7.078, time 7.22s, 
epoch 107, batches 5000|18183, train-acc 0.845, loss 0.400, para-norm 5024.740, grad-norm 6.184, time 7.22s, 
epoch 107, batches 6000|18183, train-acc 0.846, loss 0.398, para-norm 5025.417, grad-norm 4.997, time 7.23s, 
epoch 107, batches 7000|18183, train-acc 0.843, loss 0.406, para-norm 5025.602, grad-norm 2.439, time 7.22s, 
epoch 107, batches 8000|18183, train-acc 0.848, loss 0.395, para-norm 5026.043, grad-norm 2.204, time 7.22s, 
epoch 107, batches 9000|18183, train-acc 0.845, loss 0.403, para-norm 5026.131, grad-norm 6.544, time 7.23s, 
epoch 107, batches 10000|18183, train-acc 0.846, loss 0.400, para-norm 5025.966, grad-norm 4.842, time 7.22s, 
epoch 107, batches 11000|18183, train-acc 0.843, loss 0.407, para-norm 5025.277, grad-norm 2.720, time 7.23s, 
epoch 107, batches 12000|18183, train-acc 0.847, loss 0.395, para-norm 5025.235, grad-norm 4.978, time 7.21s, 
epoch 107, batches 13000|18183, train-acc 0.845, loss 0.401, para-norm 5025.182, grad-norm 2.492, time 7.24s, 
epoch 107, batches 14000|18183, train-acc 0.848, loss 0.394, para-norm 5025.159, grad-norm 3.919, time 7.21s, 
epoch 107, batches 15000|18183, train-acc 0.847, loss 0.400, para-norm 5024.719, grad-norm 6.107, time 7.21s, 
epoch 107, batches 16000|18183, train-acc 0.847, loss 0.401, para-norm 5023.922, grad-norm 3.078, time 7.22s, 
epoch 107, batches 17000|18183, train-acc 0.846, loss 0.399, para-norm 5023.935, grad-norm 5.627, time 7.23s, 
epoch 107, batches 18000|18183, train-acc 0.847, loss 0.398, para-norm 5022.764, grad-norm 5.081, time 7.21s, 
epoch 107, batches 18183|18183, train-acc 0.845, loss 0.388, para-norm 5023.141, grad-norm 3.162, time 1.32s, 
dev-acc 0.855
epoch 108, batches 1000|18183, train-acc 0.847, loss 0.396, para-norm 5025.199, grad-norm 4.330, time 7.29s, 
epoch 108, batches 2000|18183, train-acc 0.847, loss 0.396, para-norm 5025.157, grad-norm 2.578, time 7.24s, 
epoch 108, batches 3000|18183, train-acc 0.850, loss 0.392, para-norm 5025.119, grad-norm 2.918, time 7.22s, 
epoch 108, batches 4000|18183, train-acc 0.842, loss 0.401, para-norm 5024.665, grad-norm 4.226, time 7.31s, 
epoch 108, batches 5000|18183, train-acc 0.848, loss 0.394, para-norm 5024.534, grad-norm 8.157, time 7.04s, 
epoch 108, batches 6000|18183, train-acc 0.846, loss 0.398, para-norm 5024.991, grad-norm 6.324, time 6.96s, 
epoch 108, batches 7000|18183, train-acc 0.845, loss 0.399, para-norm 5025.301, grad-norm 3.859, time 6.96s, 
epoch 108, batches 8000|18183, train-acc 0.847, loss 0.397, para-norm 5026.266, grad-norm 7.879, time 6.96s, 
epoch 108, batches 9000|18183, train-acc 0.848, loss 0.397, para-norm 5026.779, grad-norm 2.842, time 6.97s, 
epoch 108, batches 10000|18183, train-acc 0.848, loss 0.394, para-norm 5025.915, grad-norm 1.766, time 6.95s, 
epoch 108, batches 11000|18183, train-acc 0.847, loss 0.396, para-norm 5025.535, grad-norm 3.624, time 6.98s, 
epoch 108, batches 12000|18183, train-acc 0.846, loss 0.405, para-norm 5024.520, grad-norm 3.970, time 6.96s, 
epoch 108, batches 13000|18183, train-acc 0.842, loss 0.404, para-norm 5024.784, grad-norm 2.766, time 6.96s, 
epoch 108, batches 14000|18183, train-acc 0.848, loss 0.400, para-norm 5024.766, grad-norm 1.934, time 6.96s, 
epoch 108, batches 15000|18183, train-acc 0.842, loss 0.407, para-norm 5032.546, grad-norm 2.611, time 6.94s, 
epoch 108, batches 16000|18183, train-acc 0.846, loss 0.401, para-norm 5031.273, grad-norm 1.654, time 6.95s, 
epoch 108, batches 17000|18183, train-acc 0.848, loss 0.398, para-norm 5031.636, grad-norm 4.409, time 6.95s, 
epoch 108, batches 18000|18183, train-acc 0.847, loss 0.397, para-norm 5030.489, grad-norm 5.194, time 6.95s, 
epoch 108, batches 18183|18183, train-acc 0.838, loss 0.420, para-norm 5030.727, grad-norm 4.938, time 1.28s, 
dev-acc 0.855
epoch 109, batches 1000|18183, train-acc 0.849, loss 0.394, para-norm 5032.277, grad-norm 5.173, time 7.06s, 
epoch 109, batches 2000|18183, train-acc 0.845, loss 0.397, para-norm 5032.494, grad-norm 3.636, time 6.95s, 
epoch 109, batches 3000|18183, train-acc 0.846, loss 0.401, para-norm 5031.788, grad-norm 6.461, time 6.97s, 
epoch 109, batches 4000|18183, train-acc 0.849, loss 0.393, para-norm 5031.331, grad-norm 2.897, time 6.97s, 
epoch 109, batches 5000|18183, train-acc 0.852, loss 0.392, para-norm 5031.429, grad-norm 1.985, time 6.96s, 
epoch 109, batches 6000|18183, train-acc 0.850, loss 0.394, para-norm 5031.234, grad-norm 4.060, time 6.95s, 
epoch 109, batches 7000|18183, train-acc 0.849, loss 0.392, para-norm 5031.495, grad-norm 3.896, time 6.96s, 
epoch 109, batches 8000|18183, train-acc 0.848, loss 0.396, para-norm 5031.202, grad-norm 2.858, time 6.96s, 
epoch 109, batches 9000|18183, train-acc 0.846, loss 0.398, para-norm 5031.229, grad-norm 3.953, time 6.96s, 
epoch 109, batches 10000|18183, train-acc 0.844, loss 0.406, para-norm 5030.515, grad-norm 4.526, time 6.96s, 
epoch 109, batches 11000|18183, train-acc 0.843, loss 0.400, para-norm 5029.866, grad-norm 5.310, time 6.95s, 
epoch 109, batches 12000|18183, train-acc 0.845, loss 0.399, para-norm 5030.500, grad-norm 4.998, time 6.95s, 
epoch 109, batches 13000|18183, train-acc 0.848, loss 0.394, para-norm 5030.733, grad-norm 5.592, time 6.96s, 
epoch 109, batches 14000|18183, train-acc 0.851, loss 0.389, para-norm 5030.874, grad-norm 6.424, time 6.94s, 
epoch 109, batches 15000|18183, train-acc 0.844, loss 0.406, para-norm 5030.562, grad-norm 3.538, time 6.96s, 
epoch 109, batches 16000|18183, train-acc 0.843, loss 0.403, para-norm 5030.129, grad-norm 1.815, time 6.96s, 
epoch 109, batches 17000|18183, train-acc 0.844, loss 0.403, para-norm 5029.821, grad-norm 1.611, time 6.96s, 
epoch 109, batches 18000|18183, train-acc 0.846, loss 0.399, para-norm 5029.384, grad-norm 5.028, time 6.97s, 
epoch 109, batches 18183|18183, train-acc 0.845, loss 0.410, para-norm 5029.255, grad-norm 8.190, time 1.27s, 
dev-acc 0.857
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
save model!
epoch 110, batches 1000|18183, train-acc 0.853, loss 0.386, para-norm 5029.371, grad-norm 5.408, time 7.02s, 
epoch 110, batches 2000|18183, train-acc 0.849, loss 0.394, para-norm 5029.274, grad-norm 2.985, time 6.95s, 
epoch 110, batches 3000|18183, train-acc 0.848, loss 0.397, para-norm 5035.327, grad-norm 4.695, time 7.01s, 
epoch 110, batches 4000|18183, train-acc 0.847, loss 0.399, para-norm 5034.866, grad-norm 7.744, time 6.96s, 
epoch 110, batches 5000|18183, train-acc 0.846, loss 0.398, para-norm 5034.476, grad-norm 5.132, time 6.94s, 
epoch 110, batches 6000|18183, train-acc 0.847, loss 0.399, para-norm 5033.825, grad-norm 2.509, time 7.03s, 
epoch 110, batches 7000|18183, train-acc 0.849, loss 0.390, para-norm 5034.187, grad-norm 3.907, time 7.02s, 
epoch 110, batches 8000|18183, train-acc 0.845, loss 0.401, para-norm 5034.327, grad-norm 3.673, time 6.98s, 
epoch 110, batches 9000|18183, train-acc 0.848, loss 0.399, para-norm 5032.979, grad-norm 1.413, time 6.97s, 
epoch 110, batches 10000|18183, train-acc 0.849, loss 0.392, para-norm 5034.079, grad-norm 4.228, time 6.98s, 
epoch 110, batches 11000|18183, train-acc 0.844, loss 0.401, para-norm 5033.543, grad-norm 33.849, time 6.97s, 
epoch 110, batches 12000|18183, train-acc 0.845, loss 0.397, para-norm 5032.192, grad-norm 0.335, time 6.97s, 
epoch 110, batches 13000|18183, train-acc 0.847, loss 0.399, para-norm 5031.775, grad-norm 3.819, time 6.97s, 
epoch 110, batches 14000|18183, train-acc 0.848, loss 0.393, para-norm 5031.878, grad-norm 3.455, time 6.96s, 
epoch 110, batches 15000|18183, train-acc 0.847, loss 0.399, para-norm 5031.569, grad-norm 4.650, time 6.96s, 
epoch 110, batches 16000|18183, train-acc 0.847, loss 0.404, para-norm 5030.665, grad-norm 3.038, time 6.96s, 
epoch 110, batches 17000|18183, train-acc 0.843, loss 0.404, para-norm 5029.647, grad-norm 2.119, time 6.96s, 
epoch 110, batches 18000|18183, train-acc 0.843, loss 0.405, para-norm 5029.294, grad-norm 3.418, time 6.98s, 
epoch 110, batches 18183|18183, train-acc 0.852, loss 0.389, para-norm 5029.181, grad-norm 4.595, time 1.27s, 
dev-acc 0.856
epoch 111, batches 1000|18183, train-acc 0.849, loss 0.393, para-norm 5029.790, grad-norm 7.464, time 7.03s, 
epoch 111, batches 2000|18183, train-acc 0.850, loss 0.392, para-norm 5031.731, grad-norm 5.207, time 6.97s, 
epoch 111, batches 3000|18183, train-acc 0.848, loss 0.396, para-norm 5031.072, grad-norm 11.887, time 6.97s, 
epoch 111, batches 4000|18183, train-acc 0.846, loss 0.399, para-norm 5030.911, grad-norm 7.072, time 6.97s, 
epoch 111, batches 5000|18183, train-acc 0.850, loss 0.389, para-norm 5030.168, grad-norm 3.077, time 7.03s, 
epoch 111, batches 6000|18183, train-acc 0.849, loss 0.397, para-norm 5030.525, grad-norm 3.124, time 6.96s, 
epoch 111, batches 7000|18183, train-acc 0.846, loss 0.403, para-norm 5030.463, grad-norm 2.514, time 6.97s, 
epoch 111, batches 8000|18183, train-acc 0.850, loss 0.392, para-norm 5030.454, grad-norm 5.174, time 6.96s, 
epoch 111, batches 9000|18183, train-acc 0.845, loss 0.400, para-norm 5030.154, grad-norm 1.533, time 6.97s, 
epoch 111, batches 10000|18183, train-acc 0.845, loss 0.400, para-norm 5029.384, grad-norm 2.063, time 6.98s, 
epoch 111, batches 11000|18183, train-acc 0.847, loss 0.394, para-norm 5029.043, grad-norm 2.208, time 6.98s, 
epoch 111, batches 12000|18183, train-acc 0.846, loss 0.394, para-norm 5028.101, grad-norm 2.488, time 6.99s, 
epoch 111, batches 13000|18183, train-acc 0.845, loss 0.402, para-norm 5035.018, grad-norm 3.066, time 6.98s, 
epoch 111, batches 14000|18183, train-acc 0.849, loss 0.395, para-norm 5035.246, grad-norm 4.539, time 6.97s, 
epoch 111, batches 15000|18183, train-acc 0.849, loss 0.395, para-norm 5035.245, grad-norm 5.489, time 6.97s, 
epoch 111, batches 16000|18183, train-acc 0.851, loss 0.387, para-norm 5035.869, grad-norm 7.427, time 6.96s, 
epoch 111, batches 17000|18183, train-acc 0.850, loss 0.392, para-norm 5035.056, grad-norm 8.902, time 7.04s, 
epoch 111, batches 18000|18183, train-acc 0.845, loss 0.400, para-norm 5033.932, grad-norm 1.669, time 6.97s, 
epoch 111, batches 18183|18183, train-acc 0.847, loss 0.402, para-norm 5033.679, grad-norm 2.082, time 1.28s, 
dev-acc 0.857
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
save model!
epoch 112, batches 1000|18183, train-acc 0.849, loss 0.394, para-norm 5034.680, grad-norm 6.115, time 7.04s, 
epoch 112, batches 2000|18183, train-acc 0.847, loss 0.396, para-norm 5034.195, grad-norm 3.302, time 6.96s, 
epoch 112, batches 3000|18183, train-acc 0.846, loss 0.400, para-norm 5035.052, grad-norm 2.760, time 6.96s, 
epoch 112, batches 4000|18183, train-acc 0.848, loss 0.393, para-norm 5036.782, grad-norm 5.563, time 6.97s, 
epoch 112, batches 5000|18183, train-acc 0.848, loss 0.396, para-norm 5035.699, grad-norm 6.128, time 7.00s, 
epoch 112, batches 6000|18183, train-acc 0.846, loss 0.400, para-norm 5035.180, grad-norm 2.686, time 7.08s, 
epoch 112, batches 7000|18183, train-acc 0.848, loss 0.392, para-norm 5041.710, grad-norm 3.211, time 6.96s, 
epoch 112, batches 8000|18183, train-acc 0.851, loss 0.387, para-norm 5042.597, grad-norm 2.099, time 6.96s, 
epoch 112, batches 9000|18183, train-acc 0.846, loss 0.404, para-norm 5041.212, grad-norm 3.792, time 6.97s, 
epoch 112, batches 10000|18183, train-acc 0.844, loss 0.402, para-norm 5041.980, grad-norm 6.195, time 6.96s, 
epoch 112, batches 11000|18183, train-acc 0.849, loss 0.392, para-norm 5042.690, grad-norm 4.453, time 6.96s, 
epoch 112, batches 12000|18183, train-acc 0.847, loss 0.400, para-norm 5041.855, grad-norm 5.922, time 6.96s, 
epoch 112, batches 13000|18183, train-acc 0.845, loss 0.401, para-norm 5040.602, grad-norm 10.173, time 6.96s, 
epoch 112, batches 14000|18183, train-acc 0.845, loss 0.400, para-norm 5039.857, grad-norm 4.822, time 6.98s, 
epoch 112, batches 15000|18183, train-acc 0.847, loss 0.394, para-norm 5039.533, grad-norm 5.488, time 6.96s, 
epoch 112, batches 16000|18183, train-acc 0.851, loss 0.390, para-norm 5038.650, grad-norm 2.169, time 6.97s, 
epoch 112, batches 17000|18183, train-acc 0.847, loss 0.398, para-norm 5039.107, grad-norm 3.589, time 6.97s, 
epoch 112, batches 18000|18183, train-acc 0.849, loss 0.392, para-norm 5038.077, grad-norm 3.807, time 6.98s, 
epoch 112, batches 18183|18183, train-acc 0.853, loss 0.387, para-norm 5037.985, grad-norm 2.926, time 1.28s, 
dev-acc 0.857
epoch 113, batches 1000|18183, train-acc 0.848, loss 0.393, para-norm 5038.641, grad-norm 4.317, time 7.06s, 
epoch 113, batches 2000|18183, train-acc 0.846, loss 0.400, para-norm 5037.786, grad-norm 2.694, time 6.97s, 
epoch 113, batches 3000|18183, train-acc 0.850, loss 0.393, para-norm 5037.377, grad-norm 4.726, time 6.96s, 
epoch 113, batches 4000|18183, train-acc 0.850, loss 0.392, para-norm 5038.379, grad-norm 2.956, time 6.97s, 
epoch 113, batches 5000|18183, train-acc 0.847, loss 0.397, para-norm 5038.736, grad-norm 6.231, time 6.97s, 
epoch 113, batches 6000|18183, train-acc 0.849, loss 0.396, para-norm 5038.427, grad-norm 7.298, time 6.97s, 
epoch 113, batches 7000|18183, train-acc 0.850, loss 0.390, para-norm 5038.578, grad-norm 0.569, time 6.97s, 
epoch 113, batches 8000|18183, train-acc 0.849, loss 0.391, para-norm 5037.516, grad-norm 7.695, time 6.96s, 
epoch 113, batches 9000|18183, train-acc 0.848, loss 0.398, para-norm 5037.768, grad-norm 3.475, time 6.97s, 
epoch 113, batches 10000|18183, train-acc 0.844, loss 0.402, para-norm 5037.439, grad-norm 3.344, time 6.98s, 
epoch 113, batches 11000|18183, train-acc 0.850, loss 0.388, para-norm 5037.971, grad-norm 5.831, time 6.96s, 
epoch 113, batches 12000|18183, train-acc 0.848, loss 0.396, para-norm 5037.494, grad-norm 6.130, time 6.97s, 
epoch 113, batches 13000|18183, train-acc 0.848, loss 0.395, para-norm 5036.963, grad-norm 2.887, time 6.97s, 
epoch 113, batches 14000|18183, train-acc 0.847, loss 0.396, para-norm 5037.208, grad-norm 5.362, time 6.96s, 
epoch 113, batches 15000|18183, train-acc 0.848, loss 0.399, para-norm 5036.549, grad-norm 7.408, time 6.97s, 
epoch 113, batches 16000|18183, train-acc 0.844, loss 0.401, para-norm 5035.800, grad-norm 2.508, time 6.98s, 
epoch 113, batches 17000|18183, train-acc 0.848, loss 0.394, para-norm 5036.562, grad-norm 5.253, time 6.97s, 
epoch 113, batches 18000|18183, train-acc 0.848, loss 0.396, para-norm 5036.443, grad-norm 4.979, time 6.98s, 
epoch 113, batches 18183|18183, train-acc 0.848, loss 0.398, para-norm 5036.244, grad-norm 3.554, time 1.27s, 
dev-acc 0.855
epoch 114, batches 1000|18183, train-acc 0.849, loss 0.395, para-norm 5036.796, grad-norm 9.310, time 7.03s, 
epoch 114, batches 2000|18183, train-acc 0.849, loss 0.393, para-norm 5036.637, grad-norm 7.214, time 6.98s, 
epoch 114, batches 3000|18183, train-acc 0.848, loss 0.397, para-norm 5036.016, grad-norm 5.628, time 6.98s, 
epoch 114, batches 4000|18183, train-acc 0.844, loss 0.405, para-norm 5035.142, grad-norm 8.279, time 6.98s, 
epoch 114, batches 5000|18183, train-acc 0.849, loss 0.390, para-norm 5034.991, grad-norm 4.228, time 6.97s, 
epoch 114, batches 6000|18183, train-acc 0.848, loss 0.395, para-norm 5036.408, grad-norm 3.374, time 6.98s, 
epoch 114, batches 7000|18183, train-acc 0.851, loss 0.391, para-norm 5036.231, grad-norm 6.075, time 6.96s, 
epoch 114, batches 8000|18183, train-acc 0.850, loss 0.391, para-norm 5036.181, grad-norm 2.273, time 6.96s, 
epoch 114, batches 9000|18183, train-acc 0.852, loss 0.386, para-norm 5035.377, grad-norm 4.605, time 6.96s, 
epoch 114, batches 10000|18183, train-acc 0.848, loss 0.394, para-norm 5035.818, grad-norm 4.410, time 6.96s, 
epoch 114, batches 11000|18183, train-acc 0.848, loss 0.396, para-norm 5035.963, grad-norm 4.485, time 7.09s, 
epoch 114, batches 12000|18183, train-acc 0.848, loss 0.395, para-norm 5035.155, grad-norm 4.672, time 6.98s, 
epoch 114, batches 13000|18183, train-acc 0.851, loss 0.392, para-norm 5035.636, grad-norm 10.382, time 6.97s, 
epoch 114, batches 14000|18183, train-acc 0.848, loss 0.394, para-norm 5035.088, grad-norm 1.103, time 6.98s, 
epoch 114, batches 15000|18183, train-acc 0.847, loss 0.395, para-norm 5035.820, grad-norm 2.708, time 6.97s, 
epoch 114, batches 16000|18183, train-acc 0.850, loss 0.396, para-norm 5034.655, grad-norm 19.586, time 6.98s, 
epoch 114, batches 17000|18183, train-acc 0.849, loss 0.396, para-norm 5034.405, grad-norm 1.393, time 7.06s, 
epoch 114, batches 18000|18183, train-acc 0.847, loss 0.395, para-norm 5035.342, grad-norm 6.851, time 6.98s, 
epoch 114, batches 18183|18183, train-acc 0.848, loss 0.390, para-norm 5035.141, grad-norm 3.075, time 1.27s, 
dev-acc 0.855
epoch 115, batches 1000|18183, train-acc 0.846, loss 0.403, para-norm 5034.351, grad-norm 3.071, time 7.09s, 
epoch 115, batches 2000|18183, train-acc 0.846, loss 0.397, para-norm 5035.259, grad-norm 5.825, time 6.97s, 
epoch 115, batches 3000|18183, train-acc 0.850, loss 0.388, para-norm 5033.895, grad-norm 1.482, time 7.56s, 
epoch 115, batches 4000|18183, train-acc 0.849, loss 0.395, para-norm 5034.585, grad-norm 6.231, time 7.80s, 
epoch 115, batches 5000|18183, train-acc 0.848, loss 0.394, para-norm 5035.182, grad-norm 11.380, time 7.80s, 
epoch 115, batches 6000|18183, train-acc 0.845, loss 0.399, para-norm 5034.676, grad-norm 4.241, time 7.80s, 
epoch 115, batches 7000|18183, train-acc 0.851, loss 0.385, para-norm 5034.446, grad-norm 7.483, time 7.80s, 
epoch 115, batches 8000|18183, train-acc 0.850, loss 0.390, para-norm 5035.119, grad-norm 2.093, time 7.79s, 
epoch 115, batches 9000|18183, train-acc 0.849, loss 0.389, para-norm 5034.631, grad-norm 2.136, time 7.80s, 
epoch 115, batches 10000|18183, train-acc 0.853, loss 0.386, para-norm 5035.461, grad-norm 2.918, time 7.79s, 
epoch 115, batches 11000|18183, train-acc 0.846, loss 0.397, para-norm 5034.747, grad-norm 2.746, time 7.81s, 
epoch 115, batches 12000|18183, train-acc 0.846, loss 0.398, para-norm 5037.846, grad-norm 5.045, time 7.80s, 
epoch 115, batches 13000|18183, train-acc 0.850, loss 0.389, para-norm 5037.860, grad-norm 2.305, time 7.79s, 
epoch 115, batches 14000|18183, train-acc 0.848, loss 0.396, para-norm 5037.507, grad-norm 3.680, time 7.80s, 
epoch 115, batches 15000|18183, train-acc 0.854, loss 0.393, para-norm 5037.598, grad-norm 2.386, time 7.79s, 
epoch 115, batches 16000|18183, train-acc 0.849, loss 0.393, para-norm 5037.012, grad-norm 6.764, time 7.79s, 
epoch 115, batches 17000|18183, train-acc 0.848, loss 0.398, para-norm 5037.498, grad-norm 4.868, time 7.79s, 
epoch 115, batches 18000|18183, train-acc 0.849, loss 0.398, para-norm 5037.882, grad-norm 3.934, time 7.81s, 
epoch 115, batches 18183|18183, train-acc 0.846, loss 0.396, para-norm 5038.504, grad-norm 0.465, time 1.43s, 
dev-acc 0.857
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
save model!
epoch 116, batches 1000|18183, train-acc 0.852, loss 0.388, para-norm 5038.363, grad-norm 1.817, time 7.88s, 
epoch 116, batches 2000|18183, train-acc 0.852, loss 0.388, para-norm 5039.032, grad-norm 3.695, time 7.80s, 
epoch 116, batches 3000|18183, train-acc 0.847, loss 0.398, para-norm 5039.789, grad-norm 4.420, time 7.82s, 
epoch 116, batches 4000|18183, train-acc 0.853, loss 0.387, para-norm 5038.981, grad-norm 6.891, time 7.79s, 
epoch 116, batches 5000|18183, train-acc 0.848, loss 0.396, para-norm 5038.708, grad-norm 12.106, time 7.80s, 
epoch 116, batches 6000|18183, train-acc 0.847, loss 0.396, para-norm 5038.364, grad-norm 1.764, time 7.81s, 
epoch 116, batches 7000|18183, train-acc 0.847, loss 0.398, para-norm 5038.053, grad-norm 2.443, time 7.80s, 
epoch 116, batches 8000|18183, train-acc 0.854, loss 0.383, para-norm 5038.182, grad-norm 3.593, time 7.81s, 
epoch 116, batches 9000|18183, train-acc 0.848, loss 0.395, para-norm 5038.488, grad-norm 4.060, time 7.81s, 
epoch 116, batches 10000|18183, train-acc 0.844, loss 0.402, para-norm 5038.386, grad-norm 56.015, time 7.81s, 
epoch 116, batches 11000|18183, train-acc 0.849, loss 0.392, para-norm 5038.518, grad-norm 6.459, time 7.78s, 
epoch 116, batches 12000|18183, train-acc 0.847, loss 0.400, para-norm 5037.856, grad-norm 3.306, time 7.80s, 
epoch 116, batches 13000|18183, train-acc 0.853, loss 0.383, para-norm 5037.849, grad-norm 4.061, time 7.80s, 
epoch 116, batches 14000|18183, train-acc 0.847, loss 0.396, para-norm 5037.349, grad-norm 2.339, time 7.82s, 
epoch 116, batches 15000|18183, train-acc 0.847, loss 0.394, para-norm 5037.303, grad-norm 8.780, time 7.81s, 
epoch 116, batches 16000|18183, train-acc 0.849, loss 0.394, para-norm 5037.789, grad-norm 3.694, time 7.79s, 
epoch 116, batches 17000|18183, train-acc 0.852, loss 0.390, para-norm 5037.417, grad-norm 1.987, time 7.81s, 
epoch 116, batches 18000|18183, train-acc 0.844, loss 0.400, para-norm 5037.791, grad-norm 4.638, time 7.80s, 
epoch 116, batches 18183|18183, train-acc 0.853, loss 0.386, para-norm 5037.724, grad-norm 3.798, time 1.42s, 
dev-acc 0.855
epoch 117, batches 1000|18183, train-acc 0.851, loss 0.385, para-norm 5038.284, grad-norm 3.151, time 7.87s, 
epoch 117, batches 2000|18183, train-acc 0.853, loss 0.389, para-norm 5048.079, grad-norm 2.609, time 7.80s, 
epoch 117, batches 3000|18183, train-acc 0.845, loss 0.403, para-norm 5048.468, grad-norm 2.263, time 7.81s, 
epoch 117, batches 4000|18183, train-acc 0.853, loss 0.386, para-norm 5048.994, grad-norm 5.911, time 7.81s, 
epoch 117, batches 5000|18183, train-acc 0.846, loss 0.398, para-norm 5047.873, grad-norm 4.163, time 7.81s, 
epoch 117, batches 6000|18183, train-acc 0.848, loss 0.398, para-norm 5047.535, grad-norm 5.441, time 7.80s, 
epoch 117, batches 7000|18183, train-acc 0.848, loss 0.392, para-norm 5046.678, grad-norm 3.106, time 7.80s, 
epoch 117, batches 8000|18183, train-acc 0.848, loss 0.390, para-norm 5055.160, grad-norm 4.243, time 7.80s, 
epoch 117, batches 9000|18183, train-acc 0.849, loss 0.394, para-norm 5055.624, grad-norm 2.030, time 7.80s, 
epoch 117, batches 10000|18183, train-acc 0.853, loss 0.383, para-norm 5060.333, grad-norm 4.437, time 7.80s, 
epoch 117, batches 11000|18183, train-acc 0.848, loss 0.397, para-norm 5059.944, grad-norm 1.340, time 7.81s, 
epoch 117, batches 12000|18183, train-acc 0.850, loss 0.390, para-norm 5060.963, grad-norm 5.199, time 7.81s, 
epoch 117, batches 13000|18183, train-acc 0.850, loss 0.390, para-norm 5059.923, grad-norm 2.060, time 7.82s, 
epoch 117, batches 14000|18183, train-acc 0.848, loss 0.390, para-norm 5058.321, grad-norm 1.821, time 7.80s, 
epoch 117, batches 15000|18183, train-acc 0.848, loss 0.401, para-norm 5057.470, grad-norm 3.882, time 7.82s, 
epoch 117, batches 16000|18183, train-acc 0.848, loss 0.391, para-norm 5056.919, grad-norm 6.936, time 7.80s, 
epoch 117, batches 17000|18183, train-acc 0.847, loss 0.396, para-norm 5055.726, grad-norm 4.644, time 7.82s, 
epoch 117, batches 18000|18183, train-acc 0.850, loss 0.386, para-norm 5055.303, grad-norm 3.947, time 7.81s, 
epoch 117, batches 18183|18183, train-acc 0.848, loss 0.388, para-norm 5054.925, grad-norm 2.834, time 1.42s, 
dev-acc 0.857
epoch 118, batches 1000|18183, train-acc 0.850, loss 0.388, para-norm 5055.406, grad-norm 4.006, time 7.86s, 
epoch 118, batches 2000|18183, train-acc 0.849, loss 0.391, para-norm 5055.900, grad-norm 5.363, time 7.80s, 
epoch 118, batches 3000|18183, train-acc 0.848, loss 0.395, para-norm 5055.324, grad-norm 5.739, time 7.81s, 
epoch 118, batches 4000|18183, train-acc 0.850, loss 0.390, para-norm 5054.868, grad-norm 3.954, time 7.80s, 
epoch 118, batches 5000|18183, train-acc 0.851, loss 0.390, para-norm 5054.526, grad-norm 7.007, time 7.59s, 
epoch 118, batches 6000|18183, train-acc 0.852, loss 0.386, para-norm 5054.909, grad-norm 6.139, time 7.04s, 
epoch 118, batches 7000|18183, train-acc 0.851, loss 0.390, para-norm 5054.807, grad-norm 11.148, time 7.07s, 
epoch 118, batches 8000|18183, train-acc 0.846, loss 0.402, para-norm 5054.187, grad-norm 4.382, time 6.99s, 
epoch 118, batches 9000|18183, train-acc 0.850, loss 0.387, para-norm 5053.333, grad-norm 9.305, time 7.09s, 
epoch 118, batches 10000|18183, train-acc 0.850, loss 0.391, para-norm 5052.852, grad-norm 1.067, time 6.97s, 
epoch 118, batches 11000|18183, train-acc 0.850, loss 0.392, para-norm 5052.389, grad-norm 7.051, time 6.97s, 
epoch 118, batches 12000|18183, train-acc 0.849, loss 0.392, para-norm 5052.984, grad-norm 4.474, time 6.97s, 
epoch 118, batches 13000|18183, train-acc 0.850, loss 0.392, para-norm 5051.996, grad-norm 1.264, time 6.98s, 
epoch 118, batches 14000|18183, train-acc 0.845, loss 0.400, para-norm 5051.881, grad-norm 5.009, time 6.98s, 
epoch 118, batches 15000|18183, train-acc 0.854, loss 0.382, para-norm 5052.268, grad-norm 75.213, time 6.97s, 
epoch 118, batches 16000|18183, train-acc 0.850, loss 0.391, para-norm 5052.179, grad-norm 2.856, time 6.97s, 
epoch 118, batches 17000|18183, train-acc 0.849, loss 0.392, para-norm 5052.423, grad-norm 0.692, time 7.04s, 
epoch 118, batches 18000|18183, train-acc 0.847, loss 0.396, para-norm 5051.669, grad-norm 4.775, time 6.97s, 
epoch 118, batches 18183|18183, train-acc 0.848, loss 0.395, para-norm 5051.501, grad-norm 5.717, time 1.26s, 
dev-acc 0.857
epoch 119, batches 1000|18183, train-acc 0.851, loss 0.391, para-norm 5051.561, grad-norm 6.820, time 7.01s, 
epoch 119, batches 2000|18183, train-acc 0.853, loss 0.384, para-norm 5051.326, grad-norm 8.091, time 6.97s, 
epoch 119, batches 3000|18183, train-acc 0.849, loss 0.392, para-norm 5050.027, grad-norm 4.661, time 6.99s, 
epoch 119, batches 4000|18183, train-acc 0.848, loss 0.395, para-norm 5057.549, grad-norm 3.953, time 6.99s, 
epoch 119, batches 5000|18183, train-acc 0.847, loss 0.391, para-norm 5058.363, grad-norm 1.732, time 6.99s, 
epoch 119, batches 6000|18183, train-acc 0.852, loss 0.386, para-norm 5060.944, grad-norm 2.943, time 6.98s, 
epoch 119, batches 7000|18183, train-acc 0.850, loss 0.387, para-norm 5060.626, grad-norm 4.198, time 6.97s, 
epoch 119, batches 8000|18183, train-acc 0.849, loss 0.390, para-norm 5059.876, grad-norm 4.715, time 6.97s, 
epoch 119, batches 9000|18183, train-acc 0.853, loss 0.388, para-norm 5060.937, grad-norm 8.510, time 6.97s, 
epoch 119, batches 10000|18183, train-acc 0.849, loss 0.392, para-norm 5059.899, grad-norm 2.287, time 6.98s, 
epoch 119, batches 11000|18183, train-acc 0.849, loss 0.393, para-norm 5059.695, grad-norm 8.730, time 6.97s, 
epoch 119, batches 12000|18183, train-acc 0.853, loss 0.386, para-norm 5059.630, grad-norm 5.321, time 6.97s, 
epoch 119, batches 13000|18183, train-acc 0.849, loss 0.392, para-norm 5059.056, grad-norm 6.455, time 6.98s, 
epoch 119, batches 14000|18183, train-acc 0.849, loss 0.397, para-norm 5058.590, grad-norm 2.564, time 6.98s, 
epoch 119, batches 15000|18183, train-acc 0.852, loss 0.387, para-norm 5058.155, grad-norm 4.719, time 6.96s, 
epoch 119, batches 16000|18183, train-acc 0.847, loss 0.398, para-norm 5057.150, grad-norm 1.460, time 6.98s, 
epoch 119, batches 17000|18183, train-acc 0.849, loss 0.394, para-norm 5056.742, grad-norm 0.619, time 6.97s, 
epoch 119, batches 18000|18183, train-acc 0.849, loss 0.391, para-norm 5056.324, grad-norm 3.756, time 6.97s, 
epoch 119, batches 18183|18183, train-acc 0.837, loss 0.413, para-norm 5055.716, grad-norm 36.660, time 1.28s, 
dev-acc 0.856
epoch 120, batches 1000|18183, train-acc 0.849, loss 0.392, para-norm 5056.070, grad-norm 2.998, time 7.06s, 
epoch 120, batches 2000|18183, train-acc 0.850, loss 0.386, para-norm 5056.088, grad-norm 6.929, time 6.98s, 
epoch 120, batches 3000|18183, train-acc 0.850, loss 0.388, para-norm 5055.916, grad-norm 3.447, time 6.97s, 
epoch 120, batches 4000|18183, train-acc 0.851, loss 0.387, para-norm 5056.483, grad-norm 3.820, time 6.98s, 
epoch 120, batches 5000|18183, train-acc 0.852, loss 0.386, para-norm 5057.388, grad-norm 1.577, time 6.98s, 
epoch 120, batches 6000|18183, train-acc 0.852, loss 0.391, para-norm 5057.117, grad-norm 3.262, time 6.96s, 
epoch 120, batches 7000|18183, train-acc 0.847, loss 0.395, para-norm 5057.826, grad-norm 4.344, time 6.97s, 
epoch 120, batches 8000|18183, train-acc 0.850, loss 0.388, para-norm 5058.092, grad-norm 5.195, time 6.96s, 
epoch 120, batches 9000|18183, train-acc 0.849, loss 0.386, para-norm 5057.316, grad-norm 8.180, time 6.97s, 
epoch 120, batches 10000|18183, train-acc 0.853, loss 0.381, para-norm 5058.574, grad-norm 5.573, time 6.98s, 
epoch 120, batches 11000|18183, train-acc 0.847, loss 0.398, para-norm 5056.992, grad-norm 2.986, time 6.96s, 
epoch 120, batches 12000|18183, train-acc 0.850, loss 0.389, para-norm 5056.363, grad-norm 7.384, time 6.98s, 
epoch 120, batches 13000|18183, train-acc 0.849, loss 0.395, para-norm 5055.673, grad-norm 4.169, time 6.98s, 
epoch 120, batches 14000|18183, train-acc 0.849, loss 0.392, para-norm 5055.068, grad-norm 3.104, time 7.00s, 
epoch 120, batches 15000|18183, train-acc 0.849, loss 0.390, para-norm 5054.895, grad-norm 1.813, time 6.98s, 
epoch 120, batches 16000|18183, train-acc 0.849, loss 0.392, para-norm 5054.533, grad-norm 6.146, time 6.98s, 
epoch 120, batches 17000|18183, train-acc 0.850, loss 0.390, para-norm 5054.313, grad-norm 4.279, time 6.97s, 
epoch 120, batches 18000|18183, train-acc 0.848, loss 0.396, para-norm 5054.225, grad-norm 5.768, time 6.97s, 
epoch 120, batches 18183|18183, train-acc 0.842, loss 0.403, para-norm 5054.530, grad-norm 4.587, time 1.28s, 
dev-acc 0.859
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
save model!
epoch 121, batches 1000|18183, train-acc 0.850, loss 0.385, para-norm 5055.919, grad-norm 0.342, time 7.05s, 
epoch 121, batches 2000|18183, train-acc 0.851, loss 0.388, para-norm 5071.201, grad-norm 2.964, time 6.97s, 
epoch 121, batches 3000|18183, train-acc 0.849, loss 0.394, para-norm 5071.088, grad-norm 3.621, time 6.98s, 
epoch 121, batches 4000|18183, train-acc 0.853, loss 0.381, para-norm 5070.349, grad-norm 5.258, time 6.96s, 
epoch 121, batches 5000|18183, train-acc 0.852, loss 0.385, para-norm 5069.247, grad-norm 7.234, time 6.98s, 
epoch 121, batches 6000|18183, train-acc 0.852, loss 0.389, para-norm 5068.223, grad-norm 7.649, time 6.98s, 
epoch 121, batches 7000|18183, train-acc 0.852, loss 0.382, para-norm 5068.630, grad-norm 4.716, time 6.97s, 
epoch 121, batches 8000|18183, train-acc 0.853, loss 0.381, para-norm 5067.868, grad-norm 6.069, time 6.97s, 
epoch 121, batches 9000|18183, train-acc 0.849, loss 0.396, para-norm 5067.631, grad-norm 2.335, time 6.98s, 
epoch 121, batches 10000|18183, train-acc 0.852, loss 0.392, para-norm 5067.704, grad-norm 1.981, time 6.98s, 
epoch 121, batches 11000|18183, train-acc 0.849, loss 0.393, para-norm 5068.008, grad-norm 10.215, time 6.98s, 
epoch 121, batches 12000|18183, train-acc 0.849, loss 0.392, para-norm 5068.410, grad-norm 1.790, time 6.99s, 
epoch 121, batches 13000|18183, train-acc 0.852, loss 0.387, para-norm 5067.638, grad-norm 2.229, time 6.99s, 
epoch 121, batches 14000|18183, train-acc 0.847, loss 0.396, para-norm 5066.716, grad-norm 1.603, time 6.98s, 
epoch 121, batches 15000|18183, train-acc 0.849, loss 0.396, para-norm 5066.431, grad-norm 10.107, time 6.98s, 
epoch 121, batches 16000|18183, train-acc 0.849, loss 0.393, para-norm 5065.892, grad-norm 2.551, time 6.98s, 
epoch 121, batches 17000|18183, train-acc 0.847, loss 0.398, para-norm 5065.009, grad-norm 4.351, time 6.97s, 
epoch 121, batches 18000|18183, train-acc 0.849, loss 0.393, para-norm 5064.918, grad-norm 3.168, time 6.97s, 
epoch 121, batches 18183|18183, train-acc 0.863, loss 0.359, para-norm 5065.037, grad-norm 4.444, time 1.27s, 
dev-acc 0.858
epoch 122, batches 1000|18183, train-acc 0.854, loss 0.382, para-norm 5064.510, grad-norm 7.266, time 7.05s, 
epoch 122, batches 2000|18183, train-acc 0.850, loss 0.392, para-norm 5064.865, grad-norm 4.834, time 6.96s, 
epoch 122, batches 3000|18183, train-acc 0.851, loss 0.388, para-norm 5065.220, grad-norm 6.804, time 6.97s, 
epoch 122, batches 4000|18183, train-acc 0.850, loss 0.391, para-norm 5065.823, grad-norm 6.443, time 6.99s, 
epoch 122, batches 5000|18183, train-acc 0.854, loss 0.386, para-norm 5064.489, grad-norm 4.333, time 6.98s, 
epoch 122, batches 6000|18183, train-acc 0.851, loss 0.387, para-norm 5064.204, grad-norm 1.897, time 6.97s, 
epoch 122, batches 7000|18183, train-acc 0.852, loss 0.387, para-norm 5064.325, grad-norm 1.348, time 6.98s, 
epoch 122, batches 8000|18183, train-acc 0.851, loss 0.388, para-norm 5064.639, grad-norm 2.004, time 6.99s, 
epoch 122, batches 9000|18183, train-acc 0.848, loss 0.393, para-norm 5063.916, grad-norm 3.414, time 6.97s, 
epoch 122, batches 10000|18183, train-acc 0.853, loss 0.387, para-norm 5065.040, grad-norm 2.906, time 7.02s, 
epoch 122, batches 11000|18183, train-acc 0.849, loss 0.392, para-norm 5067.129, grad-norm 4.100, time 6.99s, 
epoch 122, batches 12000|18183, train-acc 0.849, loss 0.390, para-norm 5067.742, grad-norm 4.003, time 6.98s, 
epoch 122, batches 13000|18183, train-acc 0.857, loss 0.378, para-norm 5074.722, grad-norm 5.792, time 6.97s, 
epoch 122, batches 14000|18183, train-acc 0.848, loss 0.396, para-norm 5074.848, grad-norm 5.658, time 6.99s, 
epoch 122, batches 15000|18183, train-acc 0.850, loss 0.385, para-norm 5075.465, grad-norm 4.366, time 6.99s, 
epoch 122, batches 16000|18183, train-acc 0.850, loss 0.391, para-norm 5074.306, grad-norm 6.520, time 7.08s, 
epoch 122, batches 17000|18183, train-acc 0.850, loss 0.394, para-norm 5074.643, grad-norm 2.578, time 6.98s, 
epoch 122, batches 18000|18183, train-acc 0.849, loss 0.393, para-norm 5074.107, grad-norm 5.597, time 6.97s, 
epoch 122, batches 18183|18183, train-acc 0.843, loss 0.399, para-norm 5074.339, grad-norm 2.887, time 1.28s, 
dev-acc 0.858
epoch 123, batches 1000|18183, train-acc 0.849, loss 0.390, para-norm 5073.558, grad-norm 8.253, time 7.06s, 
epoch 123, batches 2000|18183, train-acc 0.852, loss 0.387, para-norm 5074.017, grad-norm 1.561, time 6.99s, 
epoch 123, batches 3000|18183, train-acc 0.848, loss 0.390, para-norm 5073.625, grad-norm 2.860, time 6.98s, 
epoch 123, batches 4000|18183, train-acc 0.852, loss 0.388, para-norm 5073.508, grad-norm 6.420, time 6.99s, 
epoch 123, batches 5000|18183, train-acc 0.851, loss 0.386, para-norm 5081.976, grad-norm 4.746, time 6.99s, 
epoch 123, batches 6000|18183, train-acc 0.852, loss 0.389, para-norm 5081.624, grad-norm 3.669, time 6.99s, 
epoch 123, batches 7000|18183, train-acc 0.847, loss 0.393, para-norm 5082.011, grad-norm 9.563, time 7.00s, 
epoch 123, batches 8000|18183, train-acc 0.851, loss 0.387, para-norm 5081.897, grad-norm 8.618, time 6.99s, 
epoch 123, batches 9000|18183, train-acc 0.855, loss 0.380, para-norm 5081.173, grad-norm 3.576, time 6.98s, 
epoch 123, batches 10000|18183, train-acc 0.853, loss 0.386, para-norm 5080.814, grad-norm 5.848, time 6.98s, 
epoch 123, batches 11000|18183, train-acc 0.854, loss 0.383, para-norm 5081.007, grad-norm 5.332, time 6.98s, 
epoch 123, batches 12000|18183, train-acc 0.849, loss 0.392, para-norm 5079.840, grad-norm 3.999, time 6.99s, 
epoch 123, batches 13000|18183, train-acc 0.848, loss 0.393, para-norm 5078.562, grad-norm 2.182, time 6.99s, 
epoch 123, batches 14000|18183, train-acc 0.854, loss 0.384, para-norm 5078.626, grad-norm 3.686, time 6.97s, 
epoch 123, batches 15000|18183, train-acc 0.850, loss 0.390, para-norm 5078.704, grad-norm 6.976, time 6.97s, 
epoch 123, batches 16000|18183, train-acc 0.854, loss 0.386, para-norm 5078.764, grad-norm 3.659, time 6.99s, 
epoch 123, batches 17000|18183, train-acc 0.850, loss 0.390, para-norm 5077.907, grad-norm 3.366, time 6.98s, 
epoch 123, batches 18000|18183, train-acc 0.850, loss 0.387, para-norm 5077.651, grad-norm 1.169, time 6.99s, 
epoch 123, batches 18183|18183, train-acc 0.850, loss 0.396, para-norm 5077.527, grad-norm 9.797, time 1.28s, 
dev-acc 0.857
epoch 124, batches 1000|18183, train-acc 0.851, loss 0.386, para-norm 5077.428, grad-norm 3.981, time 7.06s, 
epoch 124, batches 2000|18183, train-acc 0.852, loss 0.384, para-norm 5077.388, grad-norm 6.996, time 6.98s, 
epoch 124, batches 3000|18183, train-acc 0.850, loss 0.388, para-norm 5076.911, grad-norm 3.027, time 6.99s, 
epoch 124, batches 4000|18183, train-acc 0.854, loss 0.383, para-norm 5086.308, grad-norm 4.804, time 6.99s, 
epoch 124, batches 5000|18183, train-acc 0.853, loss 0.380, para-norm 5086.756, grad-norm 7.574, time 6.98s, 
epoch 124, batches 6000|18183, train-acc 0.850, loss 0.387, para-norm 5085.495, grad-norm 5.496, time 6.99s, 
epoch 124, batches 7000|18183, train-acc 0.853, loss 0.387, para-norm 5085.753, grad-norm 4.105, time 6.99s, 
epoch 124, batches 8000|18183, train-acc 0.852, loss 0.385, para-norm 5085.500, grad-norm 1.051, time 7.05s, 
epoch 124, batches 9000|18183, train-acc 0.850, loss 0.389, para-norm 5085.465, grad-norm 3.551, time 6.98s, 
epoch 124, batches 10000|18183, train-acc 0.849, loss 0.390, para-norm 5085.040, grad-norm 38.170, time 6.97s, 
epoch 124, batches 11000|18183, train-acc 0.853, loss 0.384, para-norm 5084.580, grad-norm 2.852, time 6.97s, 
epoch 124, batches 12000|18183, train-acc 0.853, loss 0.382, para-norm 5085.155, grad-norm 2.894, time 6.99s, 
epoch 124, batches 13000|18183, train-acc 0.848, loss 0.393, para-norm 5084.233, grad-norm 1.471, time 6.99s, 
epoch 124, batches 14000|18183, train-acc 0.852, loss 0.388, para-norm 5084.015, grad-norm 2.606, time 6.99s, 
epoch 124, batches 15000|18183, train-acc 0.853, loss 0.386, para-norm 5083.752, grad-norm 3.620, time 6.97s, 
epoch 124, batches 16000|18183, train-acc 0.848, loss 0.395, para-norm 5083.405, grad-norm 5.166, time 6.99s, 
epoch 124, batches 17000|18183, train-acc 0.852, loss 0.387, para-norm 5082.757, grad-norm 4.486, time 6.99s, 
epoch 124, batches 18000|18183, train-acc 0.851, loss 0.390, para-norm 5082.143, grad-norm 1.685, time 6.99s, 
epoch 124, batches 18183|18183, train-acc 0.848, loss 0.395, para-norm 5082.156, grad-norm 5.244, time 1.28s, 
dev-acc 0.858
epoch 125, batches 1000|18183, train-acc 0.852, loss 0.383, para-norm 5082.716, grad-norm 3.245, time 7.05s, 
epoch 125, batches 2000|18183, train-acc 0.852, loss 0.382, para-norm 5082.768, grad-norm 2.236, time 7.00s, 
epoch 125, batches 3000|18183, train-acc 0.855, loss 0.377, para-norm 5082.547, grad-norm 2.419, time 6.97s, 
epoch 125, batches 4000|18183, train-acc 0.854, loss 0.384, para-norm 5082.543, grad-norm 0.027, time 6.98s, 
epoch 125, batches 5000|18183, train-acc 0.849, loss 0.395, para-norm 5082.211, grad-norm 6.250, time 6.98s, 
epoch 125, batches 6000|18183, train-acc 0.851, loss 0.387, para-norm 5086.547, grad-norm 4.455, time 6.98s, 
epoch 125, batches 7000|18183, train-acc 0.852, loss 0.385, para-norm 5086.409, grad-norm 5.738, time 6.97s, 
epoch 125, batches 8000|18183, train-acc 0.852, loss 0.384, para-norm 5087.067, grad-norm 6.278, time 6.99s, 
epoch 125, batches 9000|18183, train-acc 0.851, loss 0.385, para-norm 5086.771, grad-norm 5.152, time 6.99s, 
epoch 125, batches 10000|18183, train-acc 0.848, loss 0.396, para-norm 5086.516, grad-norm 8.160, time 7.01s, 
epoch 125, batches 11000|18183, train-acc 0.852, loss 0.387, para-norm 5085.945, grad-norm 4.870, time 6.98s, 
epoch 125, batches 12000|18183, train-acc 0.849, loss 0.390, para-norm 5086.118, grad-norm 2.687, time 6.99s, 
epoch 125, batches 13000|18183, train-acc 0.851, loss 0.393, para-norm 5085.591, grad-norm 3.960, time 6.99s, 
epoch 125, batches 14000|18183, train-acc 0.856, loss 0.378, para-norm 5084.541, grad-norm 67.316, time 6.98s, 
epoch 125, batches 15000|18183, train-acc 0.848, loss 0.396, para-norm 5084.123, grad-norm 2.996, time 7.11s, 
epoch 125, batches 16000|18183, train-acc 0.850, loss 0.391, para-norm 5084.192, grad-norm 5.887, time 7.23s, 
epoch 125, batches 17000|18183, train-acc 0.852, loss 0.388, para-norm 5083.769, grad-norm 0.778, time 7.42s, 
epoch 125, batches 18000|18183, train-acc 0.849, loss 0.386, para-norm 5083.458, grad-norm 3.298, time 7.88s, 
epoch 125, batches 18183|18183, train-acc 0.844, loss 0.406, para-norm 5083.218, grad-norm 4.697, time 1.43s, 
dev-acc 0.857
epoch 126, batches 1000|18183, train-acc 0.851, loss 0.382, para-norm 5083.083, grad-norm 4.973, time 7.90s, 
epoch 126, batches 2000|18183, train-acc 0.853, loss 0.383, para-norm 5083.365, grad-norm 2.249, time 7.92s, 
epoch 126, batches 3000|18183, train-acc 0.851, loss 0.389, para-norm 5083.502, grad-norm 2.898, time 7.83s, 
epoch 126, batches 4000|18183, train-acc 0.852, loss 0.385, para-norm 5084.044, grad-norm 3.211, time 7.82s, 
epoch 126, batches 5000|18183, train-acc 0.848, loss 0.397, para-norm 5083.521, grad-norm 7.449, time 7.82s, 
epoch 126, batches 6000|18183, train-acc 0.850, loss 0.391, para-norm 5083.030, grad-norm 0.381, time 7.81s, 
epoch 126, batches 7000|18183, train-acc 0.852, loss 0.380, para-norm 5088.569, grad-norm 0.982, time 7.80s, 
epoch 126, batches 8000|18183, train-acc 0.851, loss 0.389, para-norm 5090.610, grad-norm 6.217, time 7.82s, 
epoch 126, batches 9000|18183, train-acc 0.854, loss 0.382, para-norm 5090.960, grad-norm 3.599, time 7.81s, 
epoch 126, batches 10000|18183, train-acc 0.852, loss 0.387, para-norm 5090.328, grad-norm 6.677, time 7.82s, 
epoch 126, batches 11000|18183, train-acc 0.854, loss 0.385, para-norm 5090.147, grad-norm 5.036, time 7.81s, 
epoch 126, batches 12000|18183, train-acc 0.852, loss 0.381, para-norm 5090.967, grad-norm 4.718, time 7.81s, 
epoch 126, batches 13000|18183, train-acc 0.853, loss 0.383, para-norm 5090.170, grad-norm 5.045, time 7.82s, 
epoch 126, batches 14000|18183, train-acc 0.853, loss 0.380, para-norm 5089.829, grad-norm 5.551, time 7.81s, 
epoch 126, batches 15000|18183, train-acc 0.851, loss 0.387, para-norm 5088.963, grad-norm 6.565, time 7.82s, 
epoch 126, batches 16000|18183, train-acc 0.849, loss 0.396, para-norm 5088.875, grad-norm 6.659, time 7.83s, 
epoch 126, batches 17000|18183, train-acc 0.851, loss 0.386, para-norm 5088.713, grad-norm 3.638, time 7.80s, 
epoch 126, batches 18000|18183, train-acc 0.852, loss 0.386, para-norm 5088.807, grad-norm 5.469, time 7.81s, 
epoch 126, batches 18183|18183, train-acc 0.853, loss 0.401, para-norm 5088.709, grad-norm 10.435, time 1.43s, 
dev-acc 0.860
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
save model!
epoch 127, batches 1000|18183, train-acc 0.849, loss 0.394, para-norm 5088.068, grad-norm 5.059, time 7.28s, 
epoch 127, batches 2000|18183, train-acc 0.856, loss 0.376, para-norm 5088.821, grad-norm 12328.041, time 7.15s, 
epoch 127, batches 3000|18183, train-acc 0.853, loss 0.381, para-norm 5088.303, grad-norm 2.590, time 6.99s, 
epoch 127, batches 4000|18183, train-acc 0.853, loss 0.382, para-norm 5088.358, grad-norm 2.416, time 6.99s, 
epoch 127, batches 5000|18183, train-acc 0.854, loss 0.383, para-norm 5091.969, grad-norm 1.214, time 6.99s, 
epoch 127, batches 6000|18183, train-acc 0.852, loss 0.385, para-norm 5093.378, grad-norm 4.433, time 6.99s, 
epoch 127, batches 7000|18183, train-acc 0.847, loss 0.391, para-norm 5092.049, grad-norm 2.093, time 7.00s, 
epoch 127, batches 8000|18183, train-acc 0.851, loss 0.387, para-norm 5092.070, grad-norm 5.450, time 7.00s, 
epoch 127, batches 9000|18183, train-acc 0.856, loss 0.382, para-norm 5092.001, grad-norm 5.968, time 6.99s, 
epoch 127, batches 10000|18183, train-acc 0.855, loss 0.379, para-norm 5091.200, grad-norm 3.674, time 6.99s, 
epoch 127, batches 11000|18183, train-acc 0.854, loss 0.383, para-norm 5090.776, grad-norm 21.255, time 7.00s, 
epoch 127, batches 12000|18183, train-acc 0.854, loss 0.384, para-norm 5090.592, grad-norm 1.909, time 7.00s, 
epoch 127, batches 13000|18183, train-acc 0.846, loss 0.396, para-norm 5090.623, grad-norm 3.283, time 7.01s, 
epoch 127, batches 14000|18183, train-acc 0.847, loss 0.392, para-norm 5090.072, grad-norm 2.978, time 7.00s, 
epoch 127, batches 15000|18183, train-acc 0.852, loss 0.387, para-norm 5090.004, grad-norm 0.918, time 6.98s, 
epoch 127, batches 16000|18183, train-acc 0.850, loss 0.392, para-norm 5089.058, grad-norm 4.929, time 6.98s, 
epoch 127, batches 17000|18183, train-acc 0.850, loss 0.383, para-norm 5088.542, grad-norm 3.719, time 6.99s, 
epoch 127, batches 18000|18183, train-acc 0.855, loss 0.382, para-norm 5089.636, grad-norm 1.721, time 6.97s, 
epoch 127, batches 18183|18183, train-acc 0.844, loss 0.401, para-norm 5089.279, grad-norm 3.362, time 1.28s, 
dev-acc 0.857
epoch 128, batches 1000|18183, train-acc 0.851, loss 0.387, para-norm 5089.634, grad-norm 4.586, time 7.06s, 
epoch 128, batches 2000|18183, train-acc 0.851, loss 0.387, para-norm 5089.415, grad-norm 1.623, time 7.00s, 
epoch 128, batches 3000|18183, train-acc 0.856, loss 0.378, para-norm 5089.601, grad-norm 6.302, time 6.98s, 
epoch 128, batches 4000|18183, train-acc 0.852, loss 0.384, para-norm 5089.841, grad-norm 10.711, time 7.06s, 
epoch 128, batches 5000|18183, train-acc 0.849, loss 0.391, para-norm 5090.334, grad-norm 5.020, time 7.00s, 
epoch 128, batches 6000|18183, train-acc 0.852, loss 0.383, para-norm 5089.599, grad-norm 8.089, time 7.00s, 
epoch 128, batches 7000|18183, train-acc 0.851, loss 0.388, para-norm 5089.407, grad-norm 5.743, time 6.99s, 
epoch 128, batches 8000|18183, train-acc 0.853, loss 0.382, para-norm 5089.624, grad-norm 3.737, time 6.99s, 
epoch 128, batches 9000|18183, train-acc 0.850, loss 0.387, para-norm 5087.630, grad-norm 2.140, time 6.99s, 
epoch 128, batches 10000|18183, train-acc 0.852, loss 0.385, para-norm 5088.326, grad-norm 4.384, time 6.99s, 
epoch 128, batches 11000|18183, train-acc 0.855, loss 0.382, para-norm 5088.119, grad-norm 1.628, time 6.98s, 
epoch 128, batches 12000|18183, train-acc 0.851, loss 0.392, para-norm 5088.048, grad-norm 3.971, time 7.00s, 
epoch 128, batches 13000|18183, train-acc 0.854, loss 0.380, para-norm 5088.253, grad-norm 3.541, time 7.05s, 
epoch 128, batches 14000|18183, train-acc 0.852, loss 0.388, para-norm 5087.701, grad-norm 3.982, time 6.98s, 
epoch 128, batches 15000|18183, train-acc 0.852, loss 0.387, para-norm 5087.683, grad-norm 4.474, time 6.99s, 
epoch 128, batches 16000|18183, train-acc 0.852, loss 0.387, para-norm 5087.762, grad-norm 2.425, time 6.98s, 
epoch 128, batches 17000|18183, train-acc 0.851, loss 0.390, para-norm 5087.323, grad-norm 3.825, time 6.99s, 
epoch 128, batches 18000|18183, train-acc 0.851, loss 0.388, para-norm 5088.755, grad-norm 4.219, time 6.98s, 
epoch 128, batches 18183|18183, train-acc 0.854, loss 0.389, para-norm 5088.640, grad-norm 5.307, time 1.27s, 
dev-acc 0.857
epoch 129, batches 1000|18183, train-acc 0.858, loss 0.371, para-norm 5089.078, grad-norm 2.401, time 7.06s, 
epoch 129, batches 2000|18183, train-acc 0.857, loss 0.378, para-norm 5090.073, grad-norm 2.007, time 6.99s, 
epoch 129, batches 3000|18183, train-acc 0.852, loss 0.388, para-norm 5089.766, grad-norm 7.602, time 7.00s, 
epoch 129, batches 4000|18183, train-acc 0.851, loss 0.392, para-norm 5089.392, grad-norm 4.022, time 6.99s, 
epoch 129, batches 5000|18183, train-acc 0.850, loss 0.390, para-norm 5089.237, grad-norm 6.263, time 6.98s, 
epoch 129, batches 6000|18183, train-acc 0.854, loss 0.381, para-norm 5089.449, grad-norm 6.563, time 6.96s, 
epoch 129, batches 7000|18183, train-acc 0.854, loss 0.383, para-norm 5089.735, grad-norm 2.301, time 6.99s, 
epoch 129, batches 8000|18183, train-acc 0.855, loss 0.380, para-norm 5089.873, grad-norm 0.003, time 6.98s, 
epoch 129, batches 9000|18183, train-acc 0.856, loss 0.376, para-norm 5090.052, grad-norm 4.020, time 6.99s, 
epoch 129, batches 10000|18183, train-acc 0.851, loss 0.383, para-norm 5089.415, grad-norm 2.532, time 6.99s, 
epoch 129, batches 11000|18183, train-acc 0.854, loss 0.384, para-norm 5089.333, grad-norm 4.841, time 6.99s, 
epoch 129, batches 12000|18183, train-acc 0.853, loss 0.386, para-norm 5088.979, grad-norm 1.980, time 6.98s, 
epoch 129, batches 13000|18183, train-acc 0.855, loss 0.382, para-norm 5088.593, grad-norm 2.508, time 6.99s, 
epoch 129, batches 14000|18183, train-acc 0.853, loss 0.381, para-norm 5088.257, grad-norm 1.516, time 6.99s, 
epoch 129, batches 15000|18183, train-acc 0.849, loss 0.394, para-norm 5087.319, grad-norm 1.150, time 7.00s, 
epoch 129, batches 16000|18183, train-acc 0.852, loss 0.390, para-norm 5087.581, grad-norm 4.750, time 6.98s, 
epoch 129, batches 17000|18183, train-acc 0.853, loss 0.384, para-norm 5087.391, grad-norm 6.192, time 6.89s, 
epoch 129, batches 18000|18183, train-acc 0.851, loss 0.392, para-norm 5086.736, grad-norm 2.071, time 6.97s, 
epoch 129, batches 18183|18183, train-acc 0.847, loss 0.396, para-norm 5086.753, grad-norm 5.328, time 1.28s, 
dev-acc 0.859
epoch 130, batches 1000|18183, train-acc 0.857, loss 0.376, para-norm 5087.270, grad-norm 3.582, time 7.04s, 
epoch 130, batches 2000|18183, train-acc 0.850, loss 0.387, para-norm 5087.619, grad-norm 7.276, time 6.97s, 
epoch 130, batches 3000|18183, train-acc 0.853, loss 0.383, para-norm 5088.379, grad-norm 2.808, time 6.99s, 
epoch 130, batches 4000|18183, train-acc 0.850, loss 0.390, para-norm 5087.721, grad-norm 4.198, time 7.00s, 
epoch 130, batches 5000|18183, train-acc 0.851, loss 0.386, para-norm 5087.899, grad-norm 8.898, time 6.99s, 
epoch 130, batches 6000|18183, train-acc 0.851, loss 0.387, para-norm 5089.939, grad-norm 3.784, time 6.99s, 
epoch 130, batches 7000|18183, train-acc 0.856, loss 0.378, para-norm 5090.310, grad-norm 1.724, time 6.99s, 
epoch 130, batches 8000|18183, train-acc 0.851, loss 0.387, para-norm 5089.346, grad-norm 5.925, time 6.98s, 
epoch 130, batches 9000|18183, train-acc 0.854, loss 0.380, para-norm 5089.269, grad-norm 4.749, time 6.99s, 
epoch 130, batches 10000|18183, train-acc 0.851, loss 0.384, para-norm 5088.352, grad-norm 3.775, time 6.99s, 
epoch 130, batches 11000|18183, train-acc 0.852, loss 0.387, para-norm 5088.602, grad-norm 4.324, time 6.98s, 
epoch 130, batches 12000|18183, train-acc 0.851, loss 0.390, para-norm 5088.391, grad-norm 56.553, time 6.99s, 
epoch 130, batches 13000|18183, train-acc 0.856, loss 0.379, para-norm 5088.775, grad-norm 4.327, time 6.97s, 
epoch 130, batches 14000|18183, train-acc 0.856, loss 0.378, para-norm 5089.202, grad-norm 5.434, time 7.44s, 
epoch 130, batches 15000|18183, train-acc 0.853, loss 0.384, para-norm 5088.482, grad-norm 7.482, time 7.82s, 
epoch 130, batches 16000|18183, train-acc 0.852, loss 0.385, para-norm 5088.595, grad-norm 10.021, time 7.83s, 
epoch 130, batches 17000|18183, train-acc 0.849, loss 0.393, para-norm 5087.430, grad-norm 4.603, time 7.84s, 
epoch 130, batches 18000|18183, train-acc 0.851, loss 0.383, para-norm 5088.141, grad-norm 3.376, time 7.82s, 
epoch 130, batches 18183|18183, train-acc 0.851, loss 0.377, para-norm 5087.859, grad-norm 6.699, time 1.43s, 
dev-acc 0.858
epoch 131, batches 1000|18183, train-acc 0.851, loss 0.389, para-norm 5088.800, grad-norm 1.436, time 7.87s, 
epoch 131, batches 2000|18183, train-acc 0.854, loss 0.380, para-norm 5089.192, grad-norm 539.722, time 7.81s, 
epoch 131, batches 3000|18183, train-acc 0.855, loss 0.383, para-norm 5089.797, grad-norm 1.264, time 7.81s, 
epoch 131, batches 4000|18183, train-acc 0.857, loss 0.371, para-norm 5089.887, grad-norm 8.156, time 7.81s, 
epoch 131, batches 5000|18183, train-acc 0.851, loss 0.385, para-norm 5090.031, grad-norm 3.068, time 7.82s, 
epoch 131, batches 6000|18183, train-acc 0.854, loss 0.381, para-norm 5090.256, grad-norm 52.032, time 7.83s, 
epoch 131, batches 7000|18183, train-acc 0.853, loss 0.387, para-norm 5090.150, grad-norm 3.688, time 7.83s, 
epoch 131, batches 8000|18183, train-acc 0.853, loss 0.385, para-norm 5089.981, grad-norm 5.363, time 7.82s, 
epoch 131, batches 9000|18183, train-acc 0.853, loss 0.378, para-norm 5089.956, grad-norm 7.345, time 7.83s, 
epoch 131, batches 10000|18183, train-acc 0.852, loss 0.380, para-norm 5089.656, grad-norm 0.974, time 7.83s, 
epoch 131, batches 11000|18183, train-acc 0.850, loss 0.387, para-norm 5088.605, grad-norm 3.181, time 7.83s, 
epoch 131, batches 12000|18183, train-acc 0.851, loss 0.387, para-norm 5092.612, grad-norm 3.311, time 7.83s, 
epoch 131, batches 13000|18183, train-acc 0.849, loss 0.391, para-norm 5092.253, grad-norm 2.360, time 7.82s, 
epoch 131, batches 14000|18183, train-acc 0.853, loss 0.382, para-norm 5092.807, grad-norm 2.692, time 7.82s, 
epoch 131, batches 15000|18183, train-acc 0.851, loss 0.387, para-norm 5092.604, grad-norm 4.260, time 7.82s, 
epoch 131, batches 16000|18183, train-acc 0.850, loss 0.388, para-norm 5092.103, grad-norm 9.776, time 7.83s, 
epoch 131, batches 17000|18183, train-acc 0.854, loss 0.382, para-norm 5097.884, grad-norm 5.730, time 7.82s, 
epoch 131, batches 18000|18183, train-acc 0.853, loss 0.389, para-norm 5097.624, grad-norm 14.283, time 7.83s, 
epoch 131, batches 18183|18183, train-acc 0.856, loss 0.371, para-norm 5097.556, grad-norm 2.002, time 1.42s, 
dev-acc 0.859
epoch 132, batches 1000|18183, train-acc 0.850, loss 0.389, para-norm 5097.432, grad-norm 2.161, time 7.87s, 
epoch 132, batches 2000|18183, train-acc 0.857, loss 0.376, para-norm 5098.066, grad-norm 2.440, time 7.82s, 
epoch 132, batches 3000|18183, train-acc 0.855, loss 0.379, para-norm 5097.573, grad-norm 4.611, time 7.82s, 
epoch 132, batches 4000|18183, train-acc 0.854, loss 0.380, para-norm 5097.446, grad-norm 3.211, time 7.81s, 
epoch 132, batches 5000|18183, train-acc 0.852, loss 0.381, para-norm 5097.134, grad-norm 3.022, time 7.82s, 
epoch 132, batches 6000|18183, train-acc 0.854, loss 0.380, para-norm 5096.441, grad-norm 4.418, time 7.84s, 
epoch 132, batches 7000|18183, train-acc 0.854, loss 0.381, para-norm 5095.944, grad-norm 4.843, time 7.83s, 
epoch 132, batches 8000|18183, train-acc 0.855, loss 0.375, para-norm 5095.969, grad-norm 12.286, time 7.84s, 
epoch 132, batches 9000|18183, train-acc 0.852, loss 0.385, para-norm 5095.842, grad-norm 3.130, time 7.83s, 
epoch 132, batches 10000|18183, train-acc 0.857, loss 0.375, para-norm 5095.675, grad-norm 2.696, time 7.82s, 
epoch 132, batches 11000|18183, train-acc 0.855, loss 0.381, para-norm 5094.663, grad-norm 2.809, time 7.81s, 
epoch 132, batches 12000|18183, train-acc 0.854, loss 0.381, para-norm 5094.403, grad-norm 4.321, time 7.67s, 
epoch 132, batches 13000|18183, train-acc 0.854, loss 0.384, para-norm 5094.769, grad-norm 2.142, time 7.04s, 
epoch 132, batches 14000|18183, train-acc 0.849, loss 0.394, para-norm 5094.025, grad-norm 3.753, time 7.06s, 
epoch 132, batches 15000|18183, train-acc 0.851, loss 0.387, para-norm 5094.672, grad-norm 3.414, time 6.98s, 
epoch 132, batches 16000|18183, train-acc 0.849, loss 0.384, para-norm 5094.462, grad-norm 3.148, time 6.99s, 
epoch 132, batches 17000|18183, train-acc 0.853, loss 0.380, para-norm 5095.092, grad-norm 6.665, time 6.98s, 
epoch 132, batches 18000|18183, train-acc 0.851, loss 0.386, para-norm 5094.897, grad-norm 3.123, time 6.99s, 
epoch 132, batches 18183|18183, train-acc 0.860, loss 0.375, para-norm 5094.552, grad-norm 6.487, time 1.28s, 
dev-acc 0.858
epoch 133, batches 1000|18183, train-acc 0.854, loss 0.385, para-norm 5094.505, grad-norm 7.250, time 7.07s, 
epoch 133, batches 2000|18183, train-acc 0.856, loss 0.380, para-norm 5100.654, grad-norm 8.113, time 7.08s, 
epoch 133, batches 3000|18183, train-acc 0.856, loss 0.371, para-norm 5100.822, grad-norm 6.001, time 7.00s, 
epoch 133, batches 4000|18183, train-acc 0.854, loss 0.385, para-norm 5105.754, grad-norm 6.343, time 6.99s, 
epoch 133, batches 5000|18183, train-acc 0.856, loss 0.380, para-norm 5107.164, grad-norm 6.509, time 6.99s, 
epoch 133, batches 6000|18183, train-acc 0.854, loss 0.384, para-norm 5106.680, grad-norm 2.884, time 7.00s, 
epoch 133, batches 7000|18183, train-acc 0.854, loss 0.378, para-norm 5106.648, grad-norm 4.002, time 6.98s, 
epoch 133, batches 8000|18183, train-acc 0.857, loss 0.376, para-norm 5107.795, grad-norm 4.834, time 6.98s, 
epoch 133, batches 9000|18183, train-acc 0.855, loss 0.379, para-norm 5107.060, grad-norm 4.112, time 7.01s, 
epoch 133, batches 10000|18183, train-acc 0.854, loss 0.382, para-norm 5106.765, grad-norm 1.551, time 6.98s, 
epoch 133, batches 11000|18183, train-acc 0.853, loss 0.380, para-norm 5105.882, grad-norm 6.461, time 6.99s, 
epoch 133, batches 12000|18183, train-acc 0.853, loss 0.384, para-norm 5105.183, grad-norm 4.529, time 7.00s, 
epoch 133, batches 13000|18183, train-acc 0.852, loss 0.384, para-norm 5105.097, grad-norm 3.544, time 7.01s, 
epoch 133, batches 14000|18183, train-acc 0.853, loss 0.384, para-norm 5104.000, grad-norm 3.021, time 6.98s, 
epoch 133, batches 15000|18183, train-acc 0.853, loss 0.384, para-norm 5103.383, grad-norm 1.518, time 7.00s, 
epoch 133, batches 16000|18183, train-acc 0.852, loss 0.387, para-norm 5102.405, grad-norm 4.015, time 6.98s, 
epoch 133, batches 17000|18183, train-acc 0.854, loss 0.382, para-norm 5101.659, grad-norm 6.901, time 6.98s, 
epoch 133, batches 18000|18183, train-acc 0.850, loss 0.389, para-norm 5101.309, grad-norm 2.667, time 6.99s, 
epoch 133, batches 18183|18183, train-acc 0.851, loss 0.388, para-norm 5101.193, grad-norm 3.142, time 1.28s, 
dev-acc 0.858
epoch 134, batches 1000|18183, train-acc 0.855, loss 0.378, para-norm 5101.764, grad-norm 3.732, time 7.08s, 
epoch 134, batches 2000|18183, train-acc 0.851, loss 0.384, para-norm 5101.642, grad-norm 2183.820, time 6.99s, 
epoch 134, batches 3000|18183, train-acc 0.856, loss 0.374, para-norm 5101.648, grad-norm 2.723, time 6.99s, 
epoch 134, batches 4000|18183, train-acc 0.856, loss 0.377, para-norm 5101.417, grad-norm 7.714, time 7.00s, 
epoch 134, batches 5000|18183, train-acc 0.855, loss 0.379, para-norm 5101.487, grad-norm 2.003, time 6.99s, 
epoch 134, batches 6000|18183, train-acc 0.854, loss 0.384, para-norm 5101.175, grad-norm 2.408, time 7.00s, 
epoch 134, batches 7000|18183, train-acc 0.856, loss 0.378, para-norm 5100.822, grad-norm 5.649, time 7.00s, 
epoch 134, batches 8000|18183, train-acc 0.853, loss 0.385, para-norm 5100.104, grad-norm 6.154, time 7.00s, 
epoch 134, batches 9000|18183, train-acc 0.854, loss 0.382, para-norm 5099.348, grad-norm 3.772, time 7.08s, 
epoch 134, batches 10000|18183, train-acc 0.852, loss 0.383, para-norm 5103.506, grad-norm 6.881, time 6.99s, 
epoch 134, batches 11000|18183, train-acc 0.852, loss 0.382, para-norm 5104.209, grad-norm 2.039, time 7.00s, 
epoch 134, batches 12000|18183, train-acc 0.851, loss 0.391, para-norm 5103.250, grad-norm 3.235, time 7.01s, 
epoch 134, batches 13000|18183, train-acc 0.852, loss 0.384, para-norm 5102.600, grad-norm 3.576, time 6.99s, 
epoch 134, batches 14000|18183, train-acc 0.850, loss 0.388, para-norm 5102.535, grad-norm 6.648, time 7.01s, 
epoch 134, batches 15000|18183, train-acc 0.848, loss 0.387, para-norm 5102.040, grad-norm 4.557, time 7.00s, 
epoch 134, batches 16000|18183, train-acc 0.856, loss 0.376, para-norm 5102.349, grad-norm 0.052, time 6.98s, 
epoch 134, batches 17000|18183, train-acc 0.857, loss 0.370, para-norm 5101.913, grad-norm 2.597, time 7.00s, 
epoch 134, batches 18000|18183, train-acc 0.852, loss 0.386, para-norm 5100.762, grad-norm 5.480, time 7.01s, 
epoch 134, batches 18183|18183, train-acc 0.863, loss 0.365, para-norm 5100.635, grad-norm 7.019, time 1.28s, 
dev-acc 0.857
epoch 135, batches 1000|18183, train-acc 0.854, loss 0.378, para-norm 5101.320, grad-norm 5.184, time 7.05s, 
epoch 135, batches 2000|18183, train-acc 0.861, loss 0.367, para-norm 5101.763, grad-norm 4.615, time 6.99s, 
epoch 135, batches 3000|18183, train-acc 0.857, loss 0.376, para-norm 5102.104, grad-norm 2.044, time 7.02s, 
epoch 135, batches 4000|18183, train-acc 0.854, loss 0.382, para-norm 5100.945, grad-norm 3.804, time 7.01s, 
epoch 135, batches 5000|18183, train-acc 0.853, loss 0.385, para-norm 5101.072, grad-norm 5.557, time 7.00s, 
epoch 135, batches 6000|18183, train-acc 0.850, loss 0.387, para-norm 5101.313, grad-norm 2.082, time 6.99s, 
epoch 135, batches 7000|18183, train-acc 0.854, loss 0.383, para-norm 5101.239, grad-norm 7.668, time 7.00s, 
epoch 135, batches 8000|18183, train-acc 0.853, loss 0.380, para-norm 5102.154, grad-norm 7.685, time 7.00s, 
epoch 135, batches 9000|18183, train-acc 0.854, loss 0.382, para-norm 5101.337, grad-norm 3.701, time 7.01s, 
epoch 135, batches 10000|18183, train-acc 0.852, loss 0.387, para-norm 5100.223, grad-norm 5.994, time 7.08s, 
epoch 135, batches 11000|18183, train-acc 0.855, loss 0.381, para-norm 5100.343, grad-norm 2.197, time 6.98s, 
epoch 135, batches 12000|18183, train-acc 0.856, loss 0.377, para-norm 5100.062, grad-norm 4.638, time 6.99s, 
epoch 135, batches 13000|18183, train-acc 0.850, loss 0.389, para-norm 5099.024, grad-norm 3.640, time 6.99s, 
epoch 135, batches 14000|18183, train-acc 0.858, loss 0.374, para-norm 5099.706, grad-norm 12.457, time 6.99s, 
epoch 135, batches 15000|18183, train-acc 0.857, loss 0.378, para-norm 5099.503, grad-norm 2.218, time 7.00s, 
epoch 135, batches 16000|18183, train-acc 0.852, loss 0.384, para-norm 5098.987, grad-norm 8.270, time 6.99s, 
epoch 135, batches 17000|18183, train-acc 0.856, loss 0.377, para-norm 5099.223, grad-norm 5.509, time 6.98s, 
epoch 135, batches 18000|18183, train-acc 0.854, loss 0.386, para-norm 5098.844, grad-norm 1.960, time 7.15s, 
epoch 135, batches 18183|18183, train-acc 0.848, loss 0.400, para-norm 5098.800, grad-norm 2.150, time 1.28s, 
dev-acc 0.858
epoch 136, batches 1000|18183, train-acc 0.859, loss 0.370, para-norm 5099.474, grad-norm 3.778, time 7.16s, 
epoch 136, batches 2000|18183, train-acc 0.857, loss 0.379, para-norm 5098.862, grad-norm 2.568, time 7.00s, 
epoch 136, batches 3000|18183, train-acc 0.854, loss 0.380, para-norm 5099.039, grad-norm 7.344, time 6.97s, 
epoch 136, batches 4000|18183, train-acc 0.855, loss 0.381, para-norm 5099.143, grad-norm 1.761, time 7.00s, 
epoch 136, batches 5000|18183, train-acc 0.853, loss 0.381, para-norm 5099.889, grad-norm 5.610, time 6.99s, 
epoch 136, batches 6000|18183, train-acc 0.853, loss 0.386, para-norm 5099.418, grad-norm 0.017, time 7.00s, 
epoch 136, batches 7000|18183, train-acc 0.849, loss 0.390, para-norm 5099.375, grad-norm 2.300, time 7.01s, 
epoch 136, batches 8000|18183, train-acc 0.854, loss 0.381, para-norm 5099.299, grad-norm 8.003, time 6.99s, 
epoch 136, batches 9000|18183, train-acc 0.857, loss 0.377, para-norm 5099.606, grad-norm 1.375, time 6.99s, 
epoch 136, batches 10000|18183, train-acc 0.858, loss 0.374, para-norm 5099.217, grad-norm 6.552, time 6.98s, 
epoch 136, batches 11000|18183, train-acc 0.858, loss 0.374, para-norm 5098.762, grad-norm 2.651, time 6.99s, 
epoch 136, batches 12000|18183, train-acc 0.851, loss 0.390, para-norm 5098.042, grad-norm 5.779, time 7.01s, 
epoch 136, batches 13000|18183, train-acc 0.857, loss 0.373, para-norm 5099.089, grad-norm 4.591, time 6.98s, 
epoch 136, batches 14000|18183, train-acc 0.855, loss 0.377, para-norm 5099.600, grad-norm 60.519, time 6.99s, 
epoch 136, batches 15000|18183, train-acc 0.854, loss 0.382, para-norm 5099.930, grad-norm 3.092, time 7.00s, 
epoch 136, batches 16000|18183, train-acc 0.854, loss 0.380, para-norm 5099.180, grad-norm 8.716, time 6.99s, 
epoch 136, batches 17000|18183, train-acc 0.852, loss 0.382, para-norm 5098.127, grad-norm 3.620, time 7.00s, 
epoch 136, batches 18000|18183, train-acc 0.852, loss 0.385, para-norm 5097.819, grad-norm 0.000, time 7.00s, 
epoch 136, batches 18183|18183, train-acc 0.849, loss 0.386, para-norm 5097.839, grad-norm 1.552, time 1.27s, 
dev-acc 0.856
epoch 137, batches 1000|18183, train-acc 0.857, loss 0.372, para-norm 5098.121, grad-norm 6.509, time 7.05s, 
epoch 137, batches 2000|18183, train-acc 0.852, loss 0.386, para-norm 5098.298, grad-norm 3.453, time 7.00s, 
epoch 137, batches 3000|18183, train-acc 0.855, loss 0.378, para-norm 5098.460, grad-norm 3.556, time 7.03s, 
epoch 137, batches 4000|18183, train-acc 0.856, loss 0.373, para-norm 5097.920, grad-norm 5.658, time 7.00s, 
epoch 137, batches 5000|18183, train-acc 0.851, loss 0.388, para-norm 5097.604, grad-norm 6.511, time 6.99s, 
epoch 137, batches 6000|18183, train-acc 0.856, loss 0.377, para-norm 5096.819, grad-norm 3.474, time 7.00s, 
epoch 137, batches 7000|18183, train-acc 0.856, loss 0.377, para-norm 5097.614, grad-norm 4.811, time 6.98s, 
epoch 137, batches 8000|18183, train-acc 0.857, loss 0.377, para-norm 5097.294, grad-norm 1.735, time 6.99s, 
epoch 137, batches 9000|18183, train-acc 0.850, loss 0.388, para-norm 5097.316, grad-norm 3.959, time 7.00s, 
epoch 137, batches 10000|18183, train-acc 0.854, loss 0.381, para-norm 5097.928, grad-norm 2.677, time 7.00s, 
epoch 137, batches 11000|18183, train-acc 0.855, loss 0.379, para-norm 5098.421, grad-norm 5.601, time 6.99s, 
epoch 137, batches 12000|18183, train-acc 0.853, loss 0.380, para-norm 5099.002, grad-norm 1.949, time 6.99s, 
epoch 137, batches 13000|18183, train-acc 0.853, loss 0.381, para-norm 5098.858, grad-norm 3.535, time 7.01s, 
epoch 137, batches 14000|18183, train-acc 0.854, loss 0.378, para-norm 5098.518, grad-norm 3.735, time 6.99s, 
epoch 137, batches 15000|18183, train-acc 0.853, loss 0.380, para-norm 5098.589, grad-norm 1.895, time 6.99s, 
epoch 137, batches 16000|18183, train-acc 0.853, loss 0.379, para-norm 5098.060, grad-norm 4.806, time 6.99s, 
epoch 137, batches 17000|18183, train-acc 0.855, loss 0.378, para-norm 5098.088, grad-norm 6.205, time 6.98s, 
epoch 137, batches 18000|18183, train-acc 0.854, loss 0.380, para-norm 5097.766, grad-norm 3.554, time 7.01s, 
epoch 137, batches 18183|18183, train-acc 0.858, loss 0.372, para-norm 5097.646, grad-norm 6.609, time 1.28s, 
dev-acc 0.858
epoch 138, batches 1000|18183, train-acc 0.860, loss 0.368, para-norm 5097.900, grad-norm 5.131, time 7.05s, 
epoch 138, batches 2000|18183, train-acc 0.854, loss 0.380, para-norm 5097.794, grad-norm 4.176, time 7.00s, 
epoch 138, batches 3000|18183, train-acc 0.855, loss 0.378, para-norm 5097.197, grad-norm 6.641, time 6.99s, 
epoch 138, batches 4000|18183, train-acc 0.859, loss 0.372, para-norm 5098.120, grad-norm 2.771, time 6.99s, 
epoch 138, batches 5000|18183, train-acc 0.855, loss 0.378, para-norm 5097.613, grad-norm 3.817, time 7.00s, 
epoch 138, batches 6000|18183, train-acc 0.857, loss 0.378, para-norm 5097.318, grad-norm 6.748, time 6.97s, 
epoch 138, batches 7000|18183, train-acc 0.853, loss 0.383, para-norm 5096.779, grad-norm 2.741, time 6.99s, 
epoch 138, batches 8000|18183, train-acc 0.855, loss 0.379, para-norm 5096.648, grad-norm 4.568, time 7.00s, 
epoch 138, batches 9000|18183, train-acc 0.852, loss 0.385, para-norm 5097.147, grad-norm 3.326, time 7.01s, 
epoch 138, batches 10000|18183, train-acc 0.854, loss 0.379, para-norm 5096.419, grad-norm 4.406, time 7.00s, 
epoch 138, batches 11000|18183, train-acc 0.855, loss 0.376, para-norm 5096.698, grad-norm 4.134, time 7.00s, 
epoch 138, batches 12000|18183, train-acc 0.857, loss 0.373, para-norm 5095.674, grad-norm 1.956, time 7.04s, 
epoch 138, batches 13000|18183, train-acc 0.855, loss 0.376, para-norm 5096.451, grad-norm 4.197, time 7.00s, 
epoch 138, batches 14000|18183, train-acc 0.852, loss 0.386, para-norm 5095.809, grad-norm 4.631, time 6.98s, 
epoch 138, batches 15000|18183, train-acc 0.853, loss 0.386, para-norm 5095.897, grad-norm 3.147, time 6.99s, 
epoch 138, batches 16000|18183, train-acc 0.853, loss 0.380, para-norm 5096.318, grad-norm 1.677, time 6.99s, 
epoch 138, batches 17000|18183, train-acc 0.851, loss 0.383, para-norm 5098.070, grad-norm 3.585, time 7.01s, 
epoch 138, batches 18000|18183, train-acc 0.853, loss 0.383, para-norm 5097.999, grad-norm 4.139, time 6.99s, 
epoch 138, batches 18183|18183, train-acc 0.856, loss 0.377, para-norm 5097.979, grad-norm 5.442, time 1.28s, 
dev-acc 0.856
epoch 139, batches 1000|18183, train-acc 0.862, loss 0.363, para-norm 5097.831, grad-norm 2.237, time 7.01s, 
epoch 139, batches 2000|18183, train-acc 0.857, loss 0.375, para-norm 5097.024, grad-norm 8.229, time 6.99s, 
epoch 139, batches 3000|18183, train-acc 0.855, loss 0.381, para-norm 5097.233, grad-norm 4.962, time 6.99s, 
epoch 139, batches 4000|18183, train-acc 0.859, loss 0.367, para-norm 5097.658, grad-norm 4.442, time 6.99s, 
epoch 139, batches 5000|18183, train-acc 0.854, loss 0.383, para-norm 5097.764, grad-norm 4.377, time 7.01s, 
epoch 139, batches 6000|18183, train-acc 0.854, loss 0.382, para-norm 5097.152, grad-norm 4.768, time 6.99s, 
epoch 139, batches 7000|18183, train-acc 0.855, loss 0.379, para-norm 5096.796, grad-norm 4.463, time 7.00s, 
epoch 139, batches 8000|18183, train-acc 0.853, loss 0.380, para-norm 5097.487, grad-norm 4.813, time 7.01s, 
epoch 139, batches 9000|18183, train-acc 0.852, loss 0.381, para-norm 5096.984, grad-norm 152.793, time 6.99s, 
epoch 139, batches 10000|18183, train-acc 0.851, loss 0.388, para-norm 5097.015, grad-norm 4.566, time 6.99s, 
epoch 139, batches 11000|18183, train-acc 0.856, loss 0.379, para-norm 5097.022, grad-norm 5.083, time 7.00s, 
epoch 139, batches 12000|18183, train-acc 0.851, loss 0.384, para-norm 5097.261, grad-norm 3.671, time 6.99s, 
epoch 139, batches 13000|18183, train-acc 0.855, loss 0.378, para-norm 5096.278, grad-norm 5.804, time 6.98s, 
epoch 139, batches 14000|18183, train-acc 0.855, loss 0.378, para-norm 5096.531, grad-norm 5.723, time 6.99s, 
epoch 139, batches 15000|18183, train-acc 0.857, loss 0.377, para-norm 5096.174, grad-norm 5.188, time 7.20s, 
epoch 139, batches 16000|18183, train-acc 0.853, loss 0.381, para-norm 5095.955, grad-norm 25.926, time 7.26s, 
epoch 139, batches 17000|18183, train-acc 0.855, loss 0.384, para-norm 5096.157, grad-norm 3.861, time 7.25s, 
epoch 139, batches 18000|18183, train-acc 0.849, loss 0.390, para-norm 5102.301, grad-norm 2.701, time 7.25s, 
epoch 139, batches 18183|18183, train-acc 0.854, loss 0.384, para-norm 5102.381, grad-norm 4.285, time 1.32s, 
dev-acc 0.857
epoch 140, batches 1000|18183, train-acc 0.855, loss 0.379, para-norm 5102.797, grad-norm 4.038, time 7.33s, 
epoch 140, batches 2000|18183, train-acc 0.854, loss 0.376, para-norm 5102.856, grad-norm 3.177, time 7.23s, 
epoch 140, batches 3000|18183, train-acc 0.857, loss 0.369, para-norm 5102.792, grad-norm 4.220, time 7.24s, 
epoch 140, batches 4000|18183, train-acc 0.857, loss 0.376, para-norm 5102.582, grad-norm 14.550, time 7.25s, 
epoch 140, batches 5000|18183, train-acc 0.850, loss 0.386, para-norm 5102.543, grad-norm 0.926, time 7.27s, 
epoch 140, batches 6000|18183, train-acc 0.853, loss 0.384, para-norm 5102.290, grad-norm 5.296, time 7.23s, 
epoch 140, batches 7000|18183, train-acc 0.856, loss 0.377, para-norm 5102.883, grad-norm 5.105, time 7.26s, 
epoch 140, batches 8000|18183, train-acc 0.855, loss 0.379, para-norm 5102.601, grad-norm 2.167, time 7.24s, 
epoch 140, batches 9000|18183, train-acc 0.854, loss 0.376, para-norm 5102.259, grad-norm 6.391, time 7.25s, 
epoch 140, batches 10000|18183, train-acc 0.854, loss 0.379, para-norm 5102.615, grad-norm 8.867, time 7.26s, 
epoch 140, batches 11000|18183, train-acc 0.855, loss 0.379, para-norm 5102.404, grad-norm 4.886, time 7.26s, 
epoch 140, batches 12000|18183, train-acc 0.859, loss 0.371, para-norm 5102.036, grad-norm 8.950, time 7.25s, 
epoch 140, batches 13000|18183, train-acc 0.856, loss 0.378, para-norm 5101.487, grad-norm 9.557, time 7.24s, 
epoch 140, batches 14000|18183, train-acc 0.859, loss 0.366, para-norm 5102.147, grad-norm 3.618, time 7.25s, 
epoch 140, batches 15000|18183, train-acc 0.851, loss 0.389, para-norm 5101.783, grad-norm 1.752, time 7.26s, 
epoch 140, batches 16000|18183, train-acc 0.856, loss 0.374, para-norm 5101.439, grad-norm 5.188, time 7.25s, 
epoch 140, batches 17000|18183, train-acc 0.851, loss 0.388, para-norm 5100.797, grad-norm 0.702, time 7.26s, 
epoch 140, batches 18000|18183, train-acc 0.854, loss 0.381, para-norm 5100.884, grad-norm 2.536, time 7.25s, 
epoch 140, batches 18183|18183, train-acc 0.862, loss 0.369, para-norm 5100.786, grad-norm 7.514, time 1.32s, 
dev-acc 0.860
epoch 141, batches 1000|18183, train-acc 0.860, loss 0.374, para-norm 5100.207, grad-norm 1.758, time 7.34s, 
epoch 141, batches 2000|18183, train-acc 0.856, loss 0.373, para-norm 5100.700, grad-norm 3.815, time 7.25s, 
epoch 141, batches 3000|18183, train-acc 0.859, loss 0.374, para-norm 5101.069, grad-norm 3.078, time 7.26s, 
epoch 141, batches 4000|18183, train-acc 0.857, loss 0.374, para-norm 5100.346, grad-norm 3.460, time 7.25s, 
epoch 141, batches 5000|18183, train-acc 0.854, loss 0.381, para-norm 5099.920, grad-norm 5.721, time 7.25s, 
epoch 141, batches 6000|18183, train-acc 0.855, loss 0.377, para-norm 5100.100, grad-norm 71.277, time 7.25s, 
epoch 141, batches 7000|18183, train-acc 0.853, loss 0.385, para-norm 5100.251, grad-norm 0.000, time 7.25s, 
epoch 141, batches 8000|18183, train-acc 0.857, loss 0.375, para-norm 5100.718, grad-norm 3.695, time 7.24s, 
epoch 141, batches 9000|18183, train-acc 0.854, loss 0.382, para-norm 5100.856, grad-norm 6.138, time 7.26s, 
epoch 141, batches 10000|18183, train-acc 0.857, loss 0.375, para-norm 5101.256, grad-norm 3.814, time 7.26s, 
epoch 141, batches 11000|18183, train-acc 0.854, loss 0.379, para-norm 5100.941, grad-norm 4.325, time 7.26s, 
epoch 141, batches 12000|18183, train-acc 0.853, loss 0.381, para-norm 5101.334, grad-norm 3.298, time 7.25s, 
epoch 141, batches 13000|18183, train-acc 0.858, loss 0.372, para-norm 5101.167, grad-norm 6.925, time 7.24s, 
epoch 141, batches 14000|18183, train-acc 0.853, loss 0.382, para-norm 5101.218, grad-norm 3.307, time 7.26s, 
epoch 141, batches 15000|18183, train-acc 0.853, loss 0.384, para-norm 5100.756, grad-norm 3.309, time 7.26s, 
epoch 141, batches 16000|18183, train-acc 0.855, loss 0.378, para-norm 5100.121, grad-norm 14.483, time 7.26s, 
epoch 141, batches 17000|18183, train-acc 0.855, loss 0.378, para-norm 5099.988, grad-norm 3.141, time 7.32s, 
epoch 141, batches 18000|18183, train-acc 0.852, loss 0.384, para-norm 5099.933, grad-norm 9.729, time 7.26s, 
epoch 141, batches 18183|18183, train-acc 0.848, loss 0.381, para-norm 5100.403, grad-norm 7.612, time 1.32s, 
dev-acc 0.857
epoch 142, batches 1000|18183, train-acc 0.859, loss 0.372, para-norm 5100.297, grad-norm 3.112, time 7.29s, 
epoch 142, batches 2000|18183, train-acc 0.858, loss 0.367, para-norm 5100.162, grad-norm 2.745, time 7.26s, 
epoch 142, batches 3000|18183, train-acc 0.861, loss 0.364, para-norm 5100.324, grad-norm 9.964, time 7.32s, 
epoch 142, batches 4000|18183, train-acc 0.853, loss 0.384, para-norm 5100.461, grad-norm 1.300, time 7.10s, 
epoch 142, batches 5000|18183, train-acc 0.856, loss 0.378, para-norm 5100.175, grad-norm 4.021, time 7.20s, 
epoch 142, batches 6000|18183, train-acc 0.855, loss 0.379, para-norm 5100.334, grad-norm 0.139, time 7.26s, 
epoch 142, batches 7000|18183, train-acc 0.852, loss 0.383, para-norm 5099.966, grad-norm 5.293, time 7.00s, 
epoch 142, batches 8000|18183, train-acc 0.853, loss 0.384, para-norm 5099.714, grad-norm 6.019, time 6.99s, 
epoch 142, batches 9000|18183, train-acc 0.855, loss 0.377, para-norm 5100.093, grad-norm 6.337, time 7.00s, 
epoch 142, batches 10000|18183, train-acc 0.854, loss 0.377, para-norm 5099.873, grad-norm 0.619, time 6.99s, 
epoch 142, batches 11000|18183, train-acc 0.857, loss 0.376, para-norm 5099.717, grad-norm 2.606, time 7.01s, 
epoch 142, batches 12000|18183, train-acc 0.853, loss 0.383, para-norm 5099.418, grad-norm 2.818, time 6.98s, 
epoch 142, batches 13000|18183, train-acc 0.856, loss 0.378, para-norm 5099.109, grad-norm 4.491, time 6.98s, 
epoch 142, batches 14000|18183, train-acc 0.860, loss 0.373, para-norm 5098.382, grad-norm 7.912, time 7.22s, 
epoch 142, batches 15000|18183, train-acc 0.861, loss 0.371, para-norm 5099.179, grad-norm 5.199, time 7.83s, 
epoch 142, batches 16000|18183, train-acc 0.852, loss 0.385, para-norm 5098.302, grad-norm 6.920, time 7.86s, 
epoch 142, batches 17000|18183, train-acc 0.853, loss 0.382, para-norm 5098.598, grad-norm 4.761, time 7.84s, 
epoch 142, batches 18000|18183, train-acc 0.857, loss 0.374, para-norm 5098.462, grad-norm 7.360, time 7.84s, 
epoch 142, batches 18183|18183, train-acc 0.857, loss 0.370, para-norm 5098.624, grad-norm 2.556, time 1.44s, 
dev-acc 0.860
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
	142 0.860
save model!
epoch 143, batches 1000|18183, train-acc 0.857, loss 0.369, para-norm 5098.920, grad-norm 3.228, time 7.22s, 
epoch 143, batches 2000|18183, train-acc 0.854, loss 0.383, para-norm 5099.131, grad-norm 7.196, time 6.99s, 
epoch 143, batches 3000|18183, train-acc 0.857, loss 0.375, para-norm 5098.937, grad-norm 1.740, time 7.00s, 
epoch 143, batches 4000|18183, train-acc 0.859, loss 0.371, para-norm 5098.822, grad-norm 2.067, time 7.00s, 
epoch 143, batches 5000|18183, train-acc 0.858, loss 0.373, para-norm 5099.064, grad-norm 6.186, time 7.00s, 
epoch 143, batches 6000|18183, train-acc 0.857, loss 0.373, para-norm 5099.319, grad-norm 3.988, time 7.00s, 
epoch 143, batches 7000|18183, train-acc 0.855, loss 0.382, para-norm 5099.363, grad-norm 7.581, time 7.60s, 
epoch 143, batches 8000|18183, train-acc 0.853, loss 0.382, para-norm 5099.271, grad-norm 1.872, time 7.56s, 
epoch 143, batches 9000|18183, train-acc 0.853, loss 0.381, para-norm 5098.312, grad-norm 3.712, time 7.91s, 
epoch 143, batches 10000|18183, train-acc 0.859, loss 0.376, para-norm 5098.628, grad-norm 1.194, time 7.90s, 
epoch 143, batches 11000|18183, train-acc 0.857, loss 0.378, para-norm 5098.623, grad-norm 4.103, time 7.90s, 
epoch 143, batches 12000|18183, train-acc 0.857, loss 0.375, para-norm 5098.970, grad-norm 3.952, time 7.89s, 
epoch 143, batches 13000|18183, train-acc 0.858, loss 0.371, para-norm 5099.331, grad-norm 9.684, time 7.50s, 
epoch 143, batches 14000|18183, train-acc 0.854, loss 0.382, para-norm 5099.496, grad-norm 3.340, time 7.04s, 
epoch 143, batches 15000|18183, train-acc 0.856, loss 0.375, para-norm 5099.345, grad-norm 5.160, time 7.04s, 
epoch 143, batches 16000|18183, train-acc 0.855, loss 0.379, para-norm 5099.166, grad-norm 4.961, time 7.04s, 
epoch 143, batches 17000|18183, train-acc 0.856, loss 0.378, para-norm 5099.868, grad-norm 4.079, time 7.11s, 
epoch 143, batches 18000|18183, train-acc 0.854, loss 0.379, para-norm 5099.650, grad-norm 3.891, time 7.49s, 
epoch 143, batches 18183|18183, train-acc 0.857, loss 0.377, para-norm 5099.617, grad-norm 5.245, time 1.27s, 
dev-acc 0.860
epoch 144, batches 1000|18183, train-acc 0.857, loss 0.369, para-norm 5100.270, grad-norm 4.248, time 6.94s, 
epoch 144, batches 2000|18183, train-acc 0.854, loss 0.381, para-norm 5100.259, grad-norm 4.137, time 6.96s, 
epoch 144, batches 3000|18183, train-acc 0.858, loss 0.373, para-norm 5100.089, grad-norm 4.997, time 6.95s, 
epoch 144, batches 4000|18183, train-acc 0.859, loss 0.370, para-norm 5100.346, grad-norm 13.886, time 6.95s, 
epoch 144, batches 5000|18183, train-acc 0.857, loss 0.374, para-norm 5100.174, grad-norm 0.888, time 6.96s, 
epoch 144, batches 6000|18183, train-acc 0.859, loss 0.366, para-norm 5100.606, grad-norm 5.335, time 7.02s, 
epoch 144, batches 7000|18183, train-acc 0.858, loss 0.371, para-norm 5100.531, grad-norm 1.685, time 7.05s, 
epoch 144, batches 8000|18183, train-acc 0.856, loss 0.380, para-norm 5100.206, grad-norm 5.407, time 7.02s, 
epoch 144, batches 9000|18183, train-acc 0.852, loss 0.384, para-norm 5099.742, grad-norm 2.414, time 7.02s, 
epoch 144, batches 10000|18183, train-acc 0.857, loss 0.377, para-norm 5099.955, grad-norm 4.465, time 7.02s, 
epoch 144, batches 11000|18183, train-acc 0.857, loss 0.373, para-norm 5100.121, grad-norm 3.279, time 7.03s, 
epoch 144, batches 12000|18183, train-acc 0.854, loss 0.379, para-norm 5099.553, grad-norm 0.017, time 7.03s, 
epoch 144, batches 13000|18183, train-acc 0.856, loss 0.375, para-norm 5099.019, grad-norm 3.995, time 7.03s, 
epoch 144, batches 14000|18183, train-acc 0.852, loss 0.383, para-norm 5099.052, grad-norm 9.143, time 7.02s, 
epoch 144, batches 15000|18183, train-acc 0.855, loss 0.379, para-norm 5099.061, grad-norm 8.777, time 7.01s, 
epoch 144, batches 16000|18183, train-acc 0.851, loss 0.387, para-norm 5098.624, grad-norm 4.505, time 7.02s, 
epoch 144, batches 17000|18183, train-acc 0.852, loss 0.383, para-norm 5098.902, grad-norm 6.164, time 7.02s, 
epoch 144, batches 18000|18183, train-acc 0.857, loss 0.377, para-norm 5099.166, grad-norm 7.440, time 7.02s, 
epoch 144, batches 18183|18183, train-acc 0.860, loss 0.366, para-norm 5099.303, grad-norm 5.044, time 1.28s, 
dev-acc 0.860
epoch 145, batches 1000|18183, train-acc 0.855, loss 0.382, para-norm 5098.517, grad-norm 5.991, time 7.14s, 
epoch 145, batches 2000|18183, train-acc 0.858, loss 0.372, para-norm 5105.448, grad-norm 4.903, time 6.95s, 
epoch 145, batches 3000|18183, train-acc 0.853, loss 0.380, para-norm 5106.677, grad-norm 2.736, time 6.97s, 
epoch 145, batches 4000|18183, train-acc 0.852, loss 0.383, para-norm 5106.540, grad-norm 2.111, time 6.96s, 
epoch 145, batches 5000|18183, train-acc 0.857, loss 0.372, para-norm 5106.929, grad-norm 7.166, time 6.94s, 
epoch 145, batches 6000|18183, train-acc 0.854, loss 0.381, para-norm 5106.558, grad-norm 5.687, time 6.96s, 
epoch 145, batches 7000|18183, train-acc 0.856, loss 0.376, para-norm 5106.347, grad-norm 4.218, time 6.96s, 
epoch 145, batches 8000|18183, train-acc 0.853, loss 0.380, para-norm 5105.493, grad-norm 3.657, time 6.95s, 
epoch 145, batches 9000|18183, train-acc 0.858, loss 0.369, para-norm 5105.638, grad-norm 3.038, time 6.95s, 
epoch 145, batches 10000|18183, train-acc 0.856, loss 0.380, para-norm 5104.827, grad-norm 1.882, time 6.97s, 
epoch 145, batches 11000|18183, train-acc 0.856, loss 0.375, para-norm 5104.656, grad-norm 10.877, time 6.95s, 
epoch 145, batches 12000|18183, train-acc 0.857, loss 0.370, para-norm 5104.274, grad-norm 6.604, time 6.93s, 
epoch 145, batches 13000|18183, train-acc 0.860, loss 0.369, para-norm 5104.114, grad-norm 9.265, time 6.95s, 
epoch 145, batches 14000|18183, train-acc 0.856, loss 0.378, para-norm 5103.810, grad-norm 5.007, time 6.96s, 
epoch 145, batches 15000|18183, train-acc 0.854, loss 0.375, para-norm 5104.239, grad-norm 6.792, time 6.95s, 
epoch 145, batches 16000|18183, train-acc 0.851, loss 0.384, para-norm 5104.039, grad-norm 4.885, time 6.96s, 
epoch 145, batches 17000|18183, train-acc 0.856, loss 0.375, para-norm 5104.301, grad-norm 4.647, time 6.95s, 
epoch 145, batches 18000|18183, train-acc 0.854, loss 0.380, para-norm 5104.394, grad-norm 5.989, time 6.94s, 
epoch 145, batches 18183|18183, train-acc 0.852, loss 0.382, para-norm 5104.334, grad-norm 10.286, time 1.27s, 
dev-acc 0.858
epoch 146, batches 1000|18183, train-acc 0.859, loss 0.370, para-norm 5104.356, grad-norm 4.823, time 7.00s, 
epoch 146, batches 2000|18183, train-acc 0.857, loss 0.370, para-norm 5104.445, grad-norm 3.653, time 6.96s, 
epoch 146, batches 3000|18183, train-acc 0.858, loss 0.377, para-norm 5104.634, grad-norm 9.130, time 6.96s, 
epoch 146, batches 4000|18183, train-acc 0.857, loss 0.377, para-norm 5104.443, grad-norm 5.921, time 6.94s, 
epoch 146, batches 5000|18183, train-acc 0.857, loss 0.373, para-norm 5103.513, grad-norm 3.112, time 6.95s, 
epoch 146, batches 6000|18183, train-acc 0.854, loss 0.380, para-norm 5102.984, grad-norm 2.120, time 6.96s, 
epoch 146, batches 7000|18183, train-acc 0.859, loss 0.373, para-norm 5103.562, grad-norm 4.422, time 6.95s, 
epoch 146, batches 8000|18183, train-acc 0.857, loss 0.377, para-norm 5102.864, grad-norm 4.642, time 6.96s, 
epoch 146, batches 9000|18183, train-acc 0.861, loss 0.369, para-norm 5102.619, grad-norm 36.327, time 6.94s, 
epoch 146, batches 10000|18183, train-acc 0.853, loss 0.383, para-norm 5102.118, grad-norm 1.557, time 6.96s, 
epoch 146, batches 11000|18183, train-acc 0.857, loss 0.373, para-norm 5101.704, grad-norm 3.919, time 6.96s, 
epoch 146, batches 12000|18183, train-acc 0.854, loss 0.378, para-norm 5102.160, grad-norm 2.831, time 6.96s, 
epoch 146, batches 13000|18183, train-acc 0.861, loss 0.366, para-norm 5102.845, grad-norm 10.958, time 6.97s, 
epoch 146, batches 14000|18183, train-acc 0.861, loss 0.365, para-norm 5103.196, grad-norm 5.110, time 6.94s, 
epoch 146, batches 15000|18183, train-acc 0.852, loss 0.386, para-norm 5102.454, grad-norm 7.141, time 6.95s, 
epoch 146, batches 16000|18183, train-acc 0.852, loss 0.380, para-norm 5102.090, grad-norm 4.040, time 6.98s, 
epoch 146, batches 17000|18183, train-acc 0.856, loss 0.383, para-norm 5108.689, grad-norm 5.509, time 6.96s, 
epoch 146, batches 18000|18183, train-acc 0.858, loss 0.377, para-norm 5109.129, grad-norm 5.461, time 6.95s, 
epoch 146, batches 18183|18183, train-acc 0.855, loss 0.380, para-norm 5109.280, grad-norm 4.162, time 1.27s, 
dev-acc 0.860
epoch 147, batches 1000|18183, train-acc 0.857, loss 0.377, para-norm 5109.434, grad-norm 10.013, time 7.00s, 
epoch 147, batches 2000|18183, train-acc 0.861, loss 0.367, para-norm 5109.484, grad-norm 4.786, time 6.95s, 
epoch 147, batches 3000|18183, train-acc 0.861, loss 0.370, para-norm 5109.334, grad-norm 3.275, time 6.97s, 
epoch 147, batches 4000|18183, train-acc 0.859, loss 0.370, para-norm 5109.385, grad-norm 1.661, time 7.02s, 
epoch 147, batches 5000|18183, train-acc 0.858, loss 0.369, para-norm 5109.534, grad-norm 3.361, time 7.02s, 
epoch 147, batches 6000|18183, train-acc 0.855, loss 0.378, para-norm 5109.089, grad-norm 4.266, time 7.02s, 
epoch 147, batches 7000|18183, train-acc 0.855, loss 0.379, para-norm 5110.227, grad-norm 2.450, time 6.94s, 
epoch 147, batches 8000|18183, train-acc 0.857, loss 0.371, para-norm 5109.746, grad-norm 5.708, time 6.94s, 
epoch 147, batches 9000|18183, train-acc 0.856, loss 0.378, para-norm 5110.218, grad-norm 2.871, time 6.97s, 
epoch 147, batches 10000|18183, train-acc 0.855, loss 0.380, para-norm 5110.249, grad-norm 2.182, time 6.96s, 
epoch 147, batches 11000|18183, train-acc 0.857, loss 0.378, para-norm 5110.779, grad-norm 7.324, time 6.96s, 
epoch 147, batches 12000|18183, train-acc 0.859, loss 0.367, para-norm 5110.180, grad-norm 7.043, time 6.95s, 
epoch 147, batches 13000|18183, train-acc 0.854, loss 0.378, para-norm 5109.458, grad-norm 3.283, time 6.95s, 
epoch 147, batches 14000|18183, train-acc 0.856, loss 0.378, para-norm 5108.727, grad-norm 4.693, time 6.96s, 
epoch 147, batches 15000|18183, train-acc 0.854, loss 0.379, para-norm 5108.909, grad-norm 3.932, time 6.95s, 
epoch 147, batches 16000|18183, train-acc 0.859, loss 0.370, para-norm 5108.397, grad-norm 7.370, time 6.95s, 
epoch 147, batches 17000|18183, train-acc 0.853, loss 0.386, para-norm 5107.530, grad-norm 1.795, time 6.96s, 
epoch 147, batches 18000|18183, train-acc 0.854, loss 0.381, para-norm 5106.809, grad-norm 3.463, time 6.95s, 
epoch 147, batches 18183|18183, train-acc 0.853, loss 0.380, para-norm 5106.956, grad-norm 1.973, time 1.27s, 
dev-acc 0.862
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
	142 0.860
	147 0.862
save model!
epoch 148, batches 1000|18183, train-acc 0.861, loss 0.370, para-norm 5107.143, grad-norm 2.233, time 7.00s, 
epoch 148, batches 2000|18183, train-acc 0.860, loss 0.367, para-norm 5106.994, grad-norm 2.175, time 6.94s, 
epoch 148, batches 3000|18183, train-acc 0.858, loss 0.372, para-norm 5106.823, grad-norm 5.216, time 6.96s, 
epoch 148, batches 4000|18183, train-acc 0.855, loss 0.382, para-norm 5106.300, grad-norm 3.236, time 6.96s, 
epoch 148, batches 5000|18183, train-acc 0.860, loss 0.367, para-norm 5105.873, grad-norm 5.725, time 6.95s, 
epoch 148, batches 6000|18183, train-acc 0.860, loss 0.368, para-norm 5105.899, grad-norm 8.966, time 6.96s, 
epoch 148, batches 7000|18183, train-acc 0.860, loss 0.366, para-norm 5105.959, grad-norm 5.207, time 6.95s, 
epoch 148, batches 8000|18183, train-acc 0.854, loss 0.383, para-norm 5105.488, grad-norm 9.393, time 6.95s, 
epoch 148, batches 9000|18183, train-acc 0.856, loss 0.375, para-norm 5105.715, grad-norm 8.970, time 6.97s, 
epoch 148, batches 10000|18183, train-acc 0.856, loss 0.378, para-norm 5105.368, grad-norm 4.867, time 6.95s, 
epoch 148, batches 11000|18183, train-acc 0.855, loss 0.382, para-norm 5105.530, grad-norm 5.530, time 6.95s, 
epoch 148, batches 12000|18183, train-acc 0.861, loss 0.365, para-norm 5105.228, grad-norm 2.778, time 6.93s, 
epoch 148, batches 13000|18183, train-acc 0.860, loss 0.372, para-norm 5104.977, grad-norm 2.524, time 6.94s, 
epoch 148, batches 14000|18183, train-acc 0.858, loss 0.376, para-norm 5104.531, grad-norm 1.935, time 6.94s, 
epoch 148, batches 15000|18183, train-acc 0.855, loss 0.379, para-norm 5104.173, grad-norm 6.161, time 6.94s, 
epoch 148, batches 16000|18183, train-acc 0.851, loss 0.383, para-norm 5103.640, grad-norm 4.517, time 6.94s, 
epoch 148, batches 17000|18183, train-acc 0.857, loss 0.374, para-norm 5103.888, grad-norm 2.786, time 6.94s, 
epoch 148, batches 18000|18183, train-acc 0.856, loss 0.372, para-norm 5104.098, grad-norm 11.652, time 6.94s, 
epoch 148, batches 18183|18183, train-acc 0.853, loss 0.388, para-norm 5104.240, grad-norm 2.497, time 1.27s, 
dev-acc 0.860
epoch 149, batches 1000|18183, train-acc 0.860, loss 0.367, para-norm 5105.056, grad-norm 9.190, time 6.96s, 
epoch 149, batches 2000|18183, train-acc 0.859, loss 0.370, para-norm 5105.567, grad-norm 1.332, time 6.94s, 
epoch 149, batches 3000|18183, train-acc 0.860, loss 0.366, para-norm 5105.733, grad-norm 2.807, time 6.94s, 
epoch 149, batches 4000|18183, train-acc 0.858, loss 0.376, para-norm 5105.984, grad-norm 6.322, time 6.94s, 
epoch 149, batches 5000|18183, train-acc 0.852, loss 0.383, para-norm 5105.863, grad-norm 4.552, time 6.96s, 
epoch 149, batches 6000|18183, train-acc 0.855, loss 0.376, para-norm 5105.237, grad-norm 5.757, time 6.94s, 
epoch 149, batches 7000|18183, train-acc 0.857, loss 0.376, para-norm 5105.007, grad-norm 3.047, time 6.96s, 
epoch 149, batches 8000|18183, train-acc 0.857, loss 0.370, para-norm 5105.852, grad-norm 2.882, time 6.93s, 
epoch 149, batches 9000|18183, train-acc 0.857, loss 0.379, para-norm 5108.773, grad-norm 2.660, time 6.95s, 
epoch 149, batches 10000|18183, train-acc 0.860, loss 0.367, para-norm 5109.053, grad-norm 4.414, time 6.94s, 
epoch 149, batches 11000|18183, train-acc 0.859, loss 0.370, para-norm 5108.482, grad-norm 6.891, time 6.95s, 
epoch 149, batches 12000|18183, train-acc 0.853, loss 0.384, para-norm 5108.387, grad-norm 315.337, time 6.96s, 
epoch 149, batches 13000|18183, train-acc 0.857, loss 0.368, para-norm 5108.172, grad-norm 5.893, time 6.95s, 
epoch 149, batches 14000|18183, train-acc 0.855, loss 0.378, para-norm 5107.481, grad-norm 3.887, time 6.96s, 
epoch 149, batches 15000|18183, train-acc 0.859, loss 0.370, para-norm 5106.892, grad-norm 3.602, time 6.94s, 
epoch 149, batches 16000|18183, train-acc 0.857, loss 0.377, para-norm 5106.715, grad-norm 5.080, time 6.96s, 
epoch 149, batches 17000|18183, train-acc 0.854, loss 0.379, para-norm 5106.139, grad-norm 6.553, time 6.94s, 
epoch 149, batches 18000|18183, train-acc 0.859, loss 0.368, para-norm 5112.218, grad-norm 1.799, time 6.93s, 
epoch 149, batches 18183|18183, train-acc 0.854, loss 0.376, para-norm 5112.323, grad-norm 8.285, time 1.27s, 
dev-acc 0.859
epoch 150, batches 1000|18183, train-acc 0.858, loss 0.373, para-norm 5112.038, grad-norm 3.732, time 6.98s, 
epoch 150, batches 2000|18183, train-acc 0.857, loss 0.369, para-norm 5113.059, grad-norm 7.157, time 6.94s, 
epoch 150, batches 3000|18183, train-acc 0.858, loss 0.373, para-norm 5112.453, grad-norm 4.482, time 6.95s, 
epoch 150, batches 4000|18183, train-acc 0.856, loss 0.372, para-norm 5112.156, grad-norm 13.018, time 6.94s, 
epoch 150, batches 5000|18183, train-acc 0.859, loss 0.375, para-norm 5111.707, grad-norm 3.904, time 6.93s, 
epoch 150, batches 6000|18183, train-acc 0.857, loss 0.374, para-norm 5111.217, grad-norm 4.073, time 6.95s, 
epoch 150, batches 7000|18183, train-acc 0.857, loss 0.372, para-norm 5110.688, grad-norm 6.507, time 6.94s, 
epoch 150, batches 8000|18183, train-acc 0.852, loss 0.382, para-norm 5109.946, grad-norm 5.179, time 6.96s, 
epoch 150, batches 9000|18183, train-acc 0.862, loss 0.368, para-norm 5110.230, grad-norm 2.601, time 6.94s, 
epoch 150, batches 10000|18183, train-acc 0.862, loss 0.364, para-norm 5109.674, grad-norm 2.208, time 6.93s, 
epoch 150, batches 11000|18183, train-acc 0.859, loss 0.372, para-norm 5108.318, grad-norm 2.901, time 6.94s, 
epoch 150, batches 12000|18183, train-acc 0.855, loss 0.379, para-norm 5108.073, grad-norm 6.566, time 6.95s, 
epoch 150, batches 13000|18183, train-acc 0.857, loss 0.375, para-norm 5107.938, grad-norm 2.577, time 6.94s, 
epoch 150, batches 14000|18183, train-acc 0.857, loss 0.375, para-norm 5107.807, grad-norm 6.086, time 6.95s, 
epoch 150, batches 15000|18183, train-acc 0.856, loss 0.377, para-norm 5108.278, grad-norm 5.987, time 6.93s, 
epoch 150, batches 16000|18183, train-acc 0.854, loss 0.379, para-norm 5108.621, grad-norm 4.679, time 6.94s, 
epoch 150, batches 17000|18183, train-acc 0.861, loss 0.369, para-norm 5107.839, grad-norm 4.741, time 7.19s, 
epoch 150, batches 18000|18183, train-acc 0.852, loss 0.383, para-norm 5107.219, grad-norm 3.128, time 7.80s, 
epoch 150, batches 18183|18183, train-acc 0.858, loss 0.372, para-norm 5107.375, grad-norm 2.932, time 1.42s, 
dev-acc 0.858
epoch 151, batches 1000|18183, train-acc 0.860, loss 0.366, para-norm 5107.082, grad-norm 7.203, time 7.81s, 
epoch 151, batches 2000|18183, train-acc 0.856, loss 0.378, para-norm 5107.310, grad-norm 4.334, time 7.79s, 
epoch 151, batches 3000|18183, train-acc 0.860, loss 0.368, para-norm 5107.581, grad-norm 3.805, time 7.75s, 
epoch 151, batches 4000|18183, train-acc 0.858, loss 0.374, para-norm 5107.428, grad-norm 6.405, time 7.49s, 
epoch 151, batches 5000|18183, train-acc 0.859, loss 0.370, para-norm 5107.079, grad-norm 2.624, time 6.95s, 
epoch 151, batches 6000|18183, train-acc 0.857, loss 0.377, para-norm 5107.171, grad-norm 6.730, time 6.94s, 
epoch 151, batches 7000|18183, train-acc 0.857, loss 0.375, para-norm 5106.301, grad-norm 3.816, time 6.95s, 
epoch 151, batches 8000|18183, train-acc 0.855, loss 0.374, para-norm 5107.042, grad-norm 5.862, time 6.92s, 
epoch 151, batches 9000|18183, train-acc 0.857, loss 0.371, para-norm 5106.271, grad-norm 2.589, time 6.93s, 
epoch 151, batches 10000|18183, train-acc 0.859, loss 0.370, para-norm 5108.226, grad-norm 6.936, time 6.93s, 
epoch 151, batches 11000|18183, train-acc 0.859, loss 0.373, para-norm 5107.994, grad-norm 7.631, time 6.95s, 
epoch 151, batches 12000|18183, train-acc 0.859, loss 0.370, para-norm 5107.606, grad-norm 3.012, time 6.94s, 
epoch 151, batches 13000|18183, train-acc 0.858, loss 0.373, para-norm 5107.619, grad-norm 47.981, time 6.93s, 
epoch 151, batches 14000|18183, train-acc 0.850, loss 0.387, para-norm 5106.515, grad-norm 2.824, time 6.96s, 
epoch 151, batches 15000|18183, train-acc 0.857, loss 0.376, para-norm 5106.109, grad-norm 7.783, time 6.94s, 
epoch 151, batches 16000|18183, train-acc 0.854, loss 0.383, para-norm 5106.370, grad-norm 3.845, time 6.94s, 
epoch 151, batches 17000|18183, train-acc 0.857, loss 0.375, para-norm 5106.286, grad-norm 3.845, time 6.93s, 
epoch 151, batches 18000|18183, train-acc 0.857, loss 0.373, para-norm 5106.599, grad-norm 6.805, time 6.93s, 
epoch 151, batches 18183|18183, train-acc 0.857, loss 0.384, para-norm 5106.425, grad-norm 2.207, time 1.27s, 
dev-acc 0.859
epoch 152, batches 1000|18183, train-acc 0.859, loss 0.371, para-norm 5106.949, grad-norm 4.504, time 6.97s, 
epoch 152, batches 2000|18183, train-acc 0.856, loss 0.375, para-norm 5108.548, grad-norm 3.879, time 6.96s, 
epoch 152, batches 3000|18183, train-acc 0.854, loss 0.376, para-norm 5109.027, grad-norm 8.014, time 6.94s, 
epoch 152, batches 4000|18183, train-acc 0.857, loss 0.371, para-norm 5108.599, grad-norm 5.971, time 6.94s, 
epoch 152, batches 5000|18183, train-acc 0.852, loss 0.379, para-norm 5107.843, grad-norm 7.310, time 6.95s, 
epoch 152, batches 6000|18183, train-acc 0.860, loss 0.367, para-norm 5107.069, grad-norm 6.686, time 6.93s, 
epoch 152, batches 7000|18183, train-acc 0.855, loss 0.374, para-norm 5106.645, grad-norm 5.167, time 6.94s, 
epoch 152, batches 8000|18183, train-acc 0.859, loss 0.371, para-norm 5106.694, grad-norm 3.422, time 6.93s, 
epoch 152, batches 9000|18183, train-acc 0.861, loss 0.364, para-norm 5106.790, grad-norm 3.678, time 6.93s, 
epoch 152, batches 10000|18183, train-acc 0.854, loss 0.384, para-norm 5106.121, grad-norm 6.800, time 6.95s, 
epoch 152, batches 11000|18183, train-acc 0.855, loss 0.377, para-norm 5106.929, grad-norm 7.291, time 6.95s, 
epoch 152, batches 12000|18183, train-acc 0.858, loss 0.370, para-norm 5106.014, grad-norm 8.267, time 6.94s, 
epoch 152, batches 13000|18183, train-acc 0.858, loss 0.374, para-norm 5105.615, grad-norm 4.160, time 6.93s, 
epoch 152, batches 14000|18183, train-acc 0.855, loss 0.379, para-norm 5105.472, grad-norm 4.623, time 6.93s, 
epoch 152, batches 15000|18183, train-acc 0.858, loss 0.370, para-norm 5105.564, grad-norm 5.967, time 6.95s, 
epoch 152, batches 16000|18183, train-acc 0.859, loss 0.371, para-norm 5105.899, grad-norm 8.268, time 6.95s, 
epoch 152, batches 17000|18183, train-acc 0.858, loss 0.372, para-norm 5105.799, grad-norm 5.022, time 6.94s, 
epoch 152, batches 18000|18183, train-acc 0.858, loss 0.370, para-norm 5105.261, grad-norm 3.992, time 6.93s, 
epoch 152, batches 18183|18183, train-acc 0.857, loss 0.382, para-norm 5105.171, grad-norm 9.200, time 1.27s, 
dev-acc 0.862
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
	142 0.860
	147 0.862
	152 0.862
save model!
epoch 153, batches 1000|18183, train-acc 0.859, loss 0.367, para-norm 5111.014, grad-norm 3.752, time 6.99s, 
epoch 153, batches 2000|18183, train-acc 0.858, loss 0.371, para-norm 5111.928, grad-norm 5.941, time 6.94s, 
epoch 153, batches 3000|18183, train-acc 0.864, loss 0.363, para-norm 5112.423, grad-norm 3.721, time 6.93s, 
epoch 153, batches 4000|18183, train-acc 0.859, loss 0.374, para-norm 5112.023, grad-norm 3.745, time 6.96s, 
epoch 153, batches 5000|18183, train-acc 0.860, loss 0.367, para-norm 5111.720, grad-norm 12.551, time 6.93s, 
epoch 153, batches 6000|18183, train-acc 0.859, loss 0.369, para-norm 5111.159, grad-norm 3.448, time 6.93s, 
epoch 153, batches 7000|18183, train-acc 0.862, loss 0.367, para-norm 5111.082, grad-norm 2.163, time 6.94s, 
epoch 153, batches 8000|18183, train-acc 0.856, loss 0.378, para-norm 5111.616, grad-norm 5.545, time 7.00s, 
epoch 153, batches 9000|18183, train-acc 0.854, loss 0.376, para-norm 5111.510, grad-norm 8.496, time 6.96s, 
epoch 153, batches 10000|18183, train-acc 0.857, loss 0.372, para-norm 5111.189, grad-norm 496.812, time 6.95s, 
epoch 153, batches 11000|18183, train-acc 0.854, loss 0.379, para-norm 5110.692, grad-norm 8.690, time 6.96s, 
epoch 153, batches 12000|18183, train-acc 0.857, loss 0.373, para-norm 5110.312, grad-norm 5.724, time 6.96s, 
epoch 153, batches 13000|18183, train-acc 0.857, loss 0.374, para-norm 5110.038, grad-norm 7.190, time 6.96s, 
epoch 153, batches 14000|18183, train-acc 0.857, loss 0.377, para-norm 5109.244, grad-norm 5.962, time 6.96s, 
epoch 153, batches 15000|18183, train-acc 0.856, loss 0.379, para-norm 5108.939, grad-norm 0.786, time 6.95s, 
epoch 153, batches 16000|18183, train-acc 0.858, loss 0.375, para-norm 5109.314, grad-norm 0.000, time 7.04s, 
epoch 153, batches 17000|18183, train-acc 0.854, loss 0.381, para-norm 5109.344, grad-norm 5.408, time 6.96s, 
epoch 153, batches 18000|18183, train-acc 0.861, loss 0.369, para-norm 5109.706, grad-norm 6.801, time 6.95s, 
epoch 153, batches 18183|18183, train-acc 0.856, loss 0.372, para-norm 5109.649, grad-norm 8.699, time 1.27s, 
dev-acc 0.859
epoch 154, batches 1000|18183, train-acc 0.859, loss 0.373, para-norm 5115.126, grad-norm 1.255, time 7.00s, 
epoch 154, batches 2000|18183, train-acc 0.861, loss 0.368, para-norm 5115.814, grad-norm 4.084, time 6.94s, 
epoch 154, batches 3000|18183, train-acc 0.854, loss 0.381, para-norm 5115.325, grad-norm 3.273, time 6.95s, 
epoch 154, batches 4000|18183, train-acc 0.858, loss 0.370, para-norm 5115.877, grad-norm 3.774, time 6.94s, 
epoch 154, batches 5000|18183, train-acc 0.858, loss 0.371, para-norm 5116.354, grad-norm 6.087, time 6.96s, 
epoch 154, batches 6000|18183, train-acc 0.859, loss 0.368, para-norm 5116.417, grad-norm 4.100, time 6.95s, 
epoch 154, batches 7000|18183, train-acc 0.856, loss 0.381, para-norm 5116.223, grad-norm 18.285, time 6.96s, 
epoch 154, batches 8000|18183, train-acc 0.858, loss 0.374, para-norm 5115.269, grad-norm 1.661, time 6.95s, 
epoch 154, batches 9000|18183, train-acc 0.860, loss 0.369, para-norm 5115.208, grad-norm 0.000, time 6.95s, 
epoch 154, batches 10000|18183, train-acc 0.856, loss 0.377, para-norm 5120.254, grad-norm 4.439, time 6.96s, 
epoch 154, batches 11000|18183, train-acc 0.857, loss 0.375, para-norm 5120.190, grad-norm 5.703, time 6.96s, 
epoch 154, batches 12000|18183, train-acc 0.859, loss 0.369, para-norm 5120.530, grad-norm 11.114, time 6.94s, 
epoch 154, batches 13000|18183, train-acc 0.854, loss 0.380, para-norm 5119.338, grad-norm 2.910, time 6.95s, 
epoch 154, batches 14000|18183, train-acc 0.862, loss 0.363, para-norm 5119.468, grad-norm 0.000, time 6.95s, 
epoch 154, batches 15000|18183, train-acc 0.853, loss 0.378, para-norm 5118.283, grad-norm 2.507, time 6.96s, 
epoch 154, batches 16000|18183, train-acc 0.858, loss 0.373, para-norm 5118.191, grad-norm 0.033, time 6.94s, 
epoch 154, batches 17000|18183, train-acc 0.860, loss 0.371, para-norm 5118.707, grad-norm 6.274, time 6.92s, 
epoch 154, batches 18000|18183, train-acc 0.860, loss 0.371, para-norm 5118.227, grad-norm 5.931, time 6.94s, 
epoch 154, batches 18183|18183, train-acc 0.871, loss 0.351, para-norm 5118.117, grad-norm 2.744, time 1.27s, 
dev-acc 0.859
epoch 155, batches 1000|18183, train-acc 0.858, loss 0.369, para-norm 5117.820, grad-norm 4.681, time 6.96s, 
epoch 155, batches 2000|18183, train-acc 0.859, loss 0.369, para-norm 5118.209, grad-norm 4.897, time 7.03s, 
epoch 155, batches 3000|18183, train-acc 0.857, loss 0.372, para-norm 5118.449, grad-norm 10.460, time 6.95s, 
epoch 155, batches 4000|18183, train-acc 0.860, loss 0.367, para-norm 5118.911, grad-norm 4.480, time 7.00s, 
epoch 155, batches 5000|18183, train-acc 0.858, loss 0.372, para-norm 5118.734, grad-norm 2.939, time 6.95s, 
epoch 155, batches 6000|18183, train-acc 0.861, loss 0.361, para-norm 5119.514, grad-norm 1.313, time 6.94s, 
epoch 155, batches 7000|18183, train-acc 0.858, loss 0.373, para-norm 5119.066, grad-norm 7.742, time 6.94s, 
epoch 155, batches 8000|18183, train-acc 0.857, loss 0.373, para-norm 5118.716, grad-norm 25.018, time 6.95s, 
epoch 155, batches 9000|18183, train-acc 0.857, loss 0.372, para-norm 5118.800, grad-norm 4.092, time 6.95s, 
epoch 155, batches 10000|18183, train-acc 0.857, loss 0.374, para-norm 5118.733, grad-norm 5.731, time 6.94s, 
epoch 155, batches 11000|18183, train-acc 0.860, loss 0.372, para-norm 5118.143, grad-norm 7.411, time 6.94s, 
epoch 155, batches 12000|18183, train-acc 0.856, loss 0.380, para-norm 5116.907, grad-norm 3.612, time 6.94s, 
epoch 155, batches 13000|18183, train-acc 0.858, loss 0.378, para-norm 5116.545, grad-norm 4.919, time 6.96s, 
epoch 155, batches 14000|18183, train-acc 0.856, loss 0.376, para-norm 5116.506, grad-norm 3.751, time 6.95s, 
epoch 155, batches 15000|18183, train-acc 0.858, loss 0.375, para-norm 5116.442, grad-norm 6.798, time 6.95s, 
epoch 155, batches 16000|18183, train-acc 0.860, loss 0.370, para-norm 5124.021, grad-norm 3.933, time 6.95s, 
epoch 155, batches 17000|18183, train-acc 0.856, loss 0.374, para-norm 5123.478, grad-norm 5.183, time 6.94s, 
epoch 155, batches 18000|18183, train-acc 0.859, loss 0.371, para-norm 5123.356, grad-norm 3.860, time 6.94s, 
epoch 155, batches 18183|18183, train-acc 0.856, loss 0.377, para-norm 5123.327, grad-norm 6.678, time 1.27s, 
dev-acc 0.859
epoch 156, batches 1000|18183, train-acc 0.856, loss 0.372, para-norm 5123.679, grad-norm 2.651, time 6.97s, 
epoch 156, batches 2000|18183, train-acc 0.861, loss 0.365, para-norm 5123.208, grad-norm 1.433, time 6.95s, 
epoch 156, batches 3000|18183, train-acc 0.860, loss 0.367, para-norm 5123.148, grad-norm 10.276, time 6.93s, 
epoch 156, batches 4000|18183, train-acc 0.856, loss 0.375, para-norm 5121.854, grad-norm 2.023, time 6.96s, 
epoch 156, batches 5000|18183, train-acc 0.857, loss 0.371, para-norm 5120.917, grad-norm 9.867, time 6.95s, 
epoch 156, batches 6000|18183, train-acc 0.856, loss 0.375, para-norm 5120.965, grad-norm 6.885, time 6.96s, 
epoch 156, batches 7000|18183, train-acc 0.860, loss 0.371, para-norm 5121.171, grad-norm 2.923, time 6.96s, 
epoch 156, batches 8000|18183, train-acc 0.856, loss 0.378, para-norm 5121.549, grad-norm 5.784, time 6.95s, 
epoch 156, batches 9000|18183, train-acc 0.860, loss 0.368, para-norm 5120.916, grad-norm 3.931, time 6.94s, 
epoch 156, batches 10000|18183, train-acc 0.857, loss 0.375, para-norm 5120.976, grad-norm 4.836, time 6.96s, 
epoch 156, batches 11000|18183, train-acc 0.857, loss 0.375, para-norm 5120.703, grad-norm 7.907, time 6.94s, 
epoch 156, batches 12000|18183, train-acc 0.853, loss 0.380, para-norm 5120.928, grad-norm 1.592, time 6.94s, 
epoch 156, batches 13000|18183, train-acc 0.858, loss 0.371, para-norm 5121.244, grad-norm 1.974, time 6.95s, 
epoch 156, batches 14000|18183, train-acc 0.857, loss 0.371, para-norm 5120.550, grad-norm 7.545, time 6.98s, 
epoch 156, batches 15000|18183, train-acc 0.856, loss 0.377, para-norm 5120.621, grad-norm 3.852, time 6.98s, 
epoch 156, batches 16000|18183, train-acc 0.859, loss 0.369, para-norm 5121.374, grad-norm 4.791, time 6.98s, 
epoch 156, batches 17000|18183, train-acc 0.859, loss 0.371, para-norm 5121.030, grad-norm 1.287, time 7.08s, 
epoch 156, batches 18000|18183, train-acc 0.860, loss 0.370, para-norm 5120.413, grad-norm 0.991, time 6.99s, 
epoch 156, batches 18183|18183, train-acc 0.856, loss 0.377, para-norm 5120.413, grad-norm 4.633, time 1.28s, 
dev-acc 0.860
epoch 157, batches 1000|18183, train-acc 0.858, loss 0.374, para-norm 5120.775, grad-norm 5.666, time 7.03s, 
epoch 157, batches 2000|18183, train-acc 0.858, loss 0.370, para-norm 5120.395, grad-norm 3.930, time 6.99s, 
epoch 157, batches 3000|18183, train-acc 0.855, loss 0.378, para-norm 5120.321, grad-norm 5.264, time 6.99s, 
epoch 157, batches 4000|18183, train-acc 0.863, loss 0.364, para-norm 5121.622, grad-norm 2.931, time 7.00s, 
epoch 157, batches 5000|18183, train-acc 0.858, loss 0.374, para-norm 5120.703, grad-norm 11.845, time 7.00s, 
epoch 157, batches 6000|18183, train-acc 0.861, loss 0.366, para-norm 5120.516, grad-norm 4.578, time 6.98s, 
epoch 157, batches 7000|18183, train-acc 0.857, loss 0.374, para-norm 5119.840, grad-norm 8.477, time 6.99s, 
epoch 157, batches 8000|18183, train-acc 0.862, loss 0.365, para-norm 5119.992, grad-norm 3.183, time 6.99s, 
epoch 157, batches 9000|18183, train-acc 0.860, loss 0.367, para-norm 5120.146, grad-norm 3.348, time 6.99s, 
epoch 157, batches 10000|18183, train-acc 0.857, loss 0.373, para-norm 5119.708, grad-norm 3.753, time 7.02s, 
epoch 157, batches 11000|18183, train-acc 0.855, loss 0.375, para-norm 5119.609, grad-norm 4.289, time 6.99s, 
epoch 157, batches 12000|18183, train-acc 0.860, loss 0.366, para-norm 5119.727, grad-norm 5.494, time 6.98s, 
epoch 157, batches 13000|18183, train-acc 0.857, loss 0.374, para-norm 5119.875, grad-norm 5.776, time 7.00s, 
epoch 157, batches 14000|18183, train-acc 0.858, loss 0.373, para-norm 5119.056, grad-norm 3.590, time 6.99s, 
epoch 157, batches 15000|18183, train-acc 0.858, loss 0.374, para-norm 5118.485, grad-norm 5.662, time 7.04s, 
epoch 157, batches 16000|18183, train-acc 0.858, loss 0.369, para-norm 5118.245, grad-norm 7.497, time 7.04s, 
epoch 157, batches 17000|18183, train-acc 0.856, loss 0.376, para-norm 5117.959, grad-norm 6.384, time 7.01s, 
epoch 157, batches 18000|18183, train-acc 0.859, loss 0.375, para-norm 5117.202, grad-norm 3.317, time 7.03s, 
epoch 157, batches 18183|18183, train-acc 0.853, loss 0.392, para-norm 5116.914, grad-norm 5.126, time 1.29s, 
dev-acc 0.858
epoch 158, batches 1000|18183, train-acc 0.862, loss 0.364, para-norm 5117.336, grad-norm 1.502, time 7.06s, 
epoch 158, batches 2000|18183, train-acc 0.862, loss 0.366, para-norm 5118.468, grad-norm 2.557, time 7.04s, 
epoch 158, batches 3000|18183, train-acc 0.855, loss 0.375, para-norm 5117.383, grad-norm 4.776, time 7.02s, 
epoch 158, batches 4000|18183, train-acc 0.859, loss 0.369, para-norm 5124.789, grad-norm 49.490, time 7.03s, 
epoch 158, batches 5000|18183, train-acc 0.859, loss 0.363, para-norm 5123.888, grad-norm 5.494, time 7.02s, 
epoch 158, batches 6000|18183, train-acc 0.859, loss 0.369, para-norm 5123.972, grad-norm 10.313, time 7.05s, 
epoch 158, batches 7000|18183, train-acc 0.864, loss 0.358, para-norm 5124.622, grad-norm 5.439, time 7.03s, 
epoch 158, batches 8000|18183, train-acc 0.860, loss 0.363, para-norm 5124.237, grad-norm 8.396, time 7.04s, 
epoch 158, batches 9000|18183, train-acc 0.857, loss 0.375, para-norm 5123.896, grad-norm 4.674, time 7.03s, 
epoch 158, batches 10000|18183, train-acc 0.854, loss 0.375, para-norm 5123.414, grad-norm 4.373, time 7.06s, 
epoch 158, batches 11000|18183, train-acc 0.857, loss 0.376, para-norm 5123.051, grad-norm 3.908, time 7.04s, 
epoch 158, batches 12000|18183, train-acc 0.855, loss 0.377, para-norm 5122.628, grad-norm 8.545, time 7.02s, 
epoch 158, batches 13000|18183, train-acc 0.855, loss 0.378, para-norm 5122.605, grad-norm 563.181, time 7.04s, 
epoch 158, batches 14000|18183, train-acc 0.859, loss 0.367, para-norm 5122.503, grad-norm 8.313, time 7.05s, 
epoch 158, batches 15000|18183, train-acc 0.861, loss 0.365, para-norm 5122.478, grad-norm 15.181, time 7.04s, 
epoch 158, batches 16000|18183, train-acc 0.855, loss 0.382, para-norm 5122.475, grad-norm 6.054, time 7.04s, 
epoch 158, batches 17000|18183, train-acc 0.857, loss 0.377, para-norm 5121.958, grad-norm 2.707, time 7.04s, 
epoch 158, batches 18000|18183, train-acc 0.855, loss 0.376, para-norm 5121.381, grad-norm 6.354, time 7.04s, 
epoch 158, batches 18183|18183, train-acc 0.855, loss 0.373, para-norm 5121.648, grad-norm 22.957, time 1.29s, 
dev-acc 0.859
epoch 159, batches 1000|18183, train-acc 0.859, loss 0.368, para-norm 5121.622, grad-norm 3.830, time 7.06s, 
epoch 159, batches 2000|18183, train-acc 0.861, loss 0.365, para-norm 5122.677, grad-norm 8.233, time 7.20s, 
epoch 159, batches 3000|18183, train-acc 0.856, loss 0.377, para-norm 5122.659, grad-norm 3.537, time 7.12s, 
epoch 159, batches 4000|18183, train-acc 0.865, loss 0.360, para-norm 5122.291, grad-norm 3.553, time 7.04s, 
epoch 159, batches 5000|18183, train-acc 0.860, loss 0.364, para-norm 5122.737, grad-norm 6.102, time 7.04s, 
epoch 159, batches 6000|18183, train-acc 0.862, loss 0.366, para-norm 5122.853, grad-norm 5.915, time 7.06s, 
epoch 159, batches 7000|18183, train-acc 0.857, loss 0.368, para-norm 5122.466, grad-norm 1.684, time 7.04s, 
epoch 159, batches 8000|18183, train-acc 0.859, loss 0.371, para-norm 5121.710, grad-norm 9.104, time 7.18s, 
epoch 159, batches 9000|18183, train-acc 0.855, loss 0.373, para-norm 5121.358, grad-norm 6.513, time 6.98s, 
epoch 159, batches 10000|18183, train-acc 0.859, loss 0.365, para-norm 5121.551, grad-norm 6.729, time 6.97s, 
epoch 159, batches 11000|18183, train-acc 0.858, loss 0.372, para-norm 5120.680, grad-norm 5.310, time 6.97s, 
epoch 159, batches 12000|18183, train-acc 0.859, loss 0.368, para-norm 5120.273, grad-norm 12.332, time 6.99s, 
epoch 159, batches 13000|18183, train-acc 0.857, loss 0.376, para-norm 5119.888, grad-norm 4.769, time 7.03s, 
epoch 159, batches 14000|18183, train-acc 0.854, loss 0.379, para-norm 5119.862, grad-norm 3.957, time 7.06s, 
epoch 159, batches 15000|18183, train-acc 0.856, loss 0.374, para-norm 5119.330, grad-norm 6.720, time 7.02s, 
epoch 159, batches 16000|18183, train-acc 0.858, loss 0.371, para-norm 5120.151, grad-norm 3.928, time 7.03s, 
epoch 159, batches 17000|18183, train-acc 0.856, loss 0.376, para-norm 5120.225, grad-norm 44.257, time 7.03s, 
epoch 159, batches 18000|18183, train-acc 0.855, loss 0.379, para-norm 5119.484, grad-norm 7.644, time 7.34s, 
epoch 159, batches 18183|18183, train-acc 0.854, loss 0.373, para-norm 5119.396, grad-norm 0.548, time 1.42s, 
dev-acc 0.859
epoch 160, batches 1000|18183, train-acc 0.857, loss 0.370, para-norm 5119.819, grad-norm 8.017, time 7.74s, 
epoch 160, batches 2000|18183, train-acc 0.858, loss 0.367, para-norm 5127.263, grad-norm 0.001, time 7.92s, 
epoch 160, batches 3000|18183, train-acc 0.862, loss 0.362, para-norm 5127.962, grad-norm 7.791, time 7.81s, 
epoch 160, batches 4000|18183, train-acc 0.858, loss 0.371, para-norm 5127.248, grad-norm 3.360, time 7.59s, 
epoch 160, batches 5000|18183, train-acc 0.859, loss 0.367, para-norm 5127.129, grad-norm 4.641, time 6.98s, 
epoch 160, batches 6000|18183, train-acc 0.858, loss 0.372, para-norm 5126.777, grad-norm 6.189, time 6.98s, 
epoch 160, batches 7000|18183, train-acc 0.861, loss 0.364, para-norm 5125.893, grad-norm 4.179, time 6.99s, 
epoch 160, batches 8000|18183, train-acc 0.859, loss 0.369, para-norm 5124.913, grad-norm 11.264, time 6.98s, 
epoch 160, batches 9000|18183, train-acc 0.859, loss 0.364, para-norm 5124.893, grad-norm 5.370, time 6.98s, 
epoch 160, batches 10000|18183, train-acc 0.860, loss 0.372, para-norm 5124.767, grad-norm 6.934, time 7.00s, 
epoch 160, batches 11000|18183, train-acc 0.863, loss 0.361, para-norm 5124.522, grad-norm 2.971, time 7.00s, 
epoch 160, batches 12000|18183, train-acc 0.860, loss 0.369, para-norm 5123.939, grad-norm 9.800, time 7.11s, 
epoch 160, batches 13000|18183, train-acc 0.859, loss 0.369, para-norm 5123.909, grad-norm 6.093, time 6.95s, 
epoch 160, batches 14000|18183, train-acc 0.854, loss 0.382, para-norm 5123.746, grad-norm 4.010, time 6.94s, 
epoch 160, batches 15000|18183, train-acc 0.861, loss 0.365, para-norm 5123.752, grad-norm 9.220, time 7.02s, 
epoch 160, batches 16000|18183, train-acc 0.859, loss 0.376, para-norm 5123.227, grad-norm 9.215, time 7.02s, 
epoch 160, batches 17000|18183, train-acc 0.858, loss 0.367, para-norm 5123.187, grad-norm 3.054, time 7.06s, 
epoch 160, batches 18000|18183, train-acc 0.855, loss 0.381, para-norm 5122.962, grad-norm 2.733, time 7.07s, 
epoch 160, batches 18183|18183, train-acc 0.853, loss 0.388, para-norm 5122.691, grad-norm 4.692, time 1.28s, 
dev-acc 0.860
epoch 161, batches 1000|18183, train-acc 0.864, loss 0.359, para-norm 5123.306, grad-norm 4.085, time 7.14s, 
epoch 161, batches 2000|18183, train-acc 0.860, loss 0.368, para-norm 5123.112, grad-norm 4.896, time 7.07s, 
epoch 161, batches 3000|18183, train-acc 0.857, loss 0.375, para-norm 5123.243, grad-norm 2.217, time 7.07s, 
epoch 161, batches 4000|18183, train-acc 0.859, loss 0.370, para-norm 5123.005, grad-norm 2.140, time 7.06s, 
epoch 161, batches 5000|18183, train-acc 0.859, loss 0.371, para-norm 5122.752, grad-norm 8.931, time 7.06s, 
epoch 161, batches 6000|18183, train-acc 0.858, loss 0.369, para-norm 5122.730, grad-norm 2.120, time 7.07s, 
epoch 161, batches 7000|18183, train-acc 0.861, loss 0.361, para-norm 5123.270, grad-norm 3.848, time 7.06s, 
epoch 161, batches 8000|18183, train-acc 0.859, loss 0.373, para-norm 5122.999, grad-norm 3.765, time 7.07s, 
epoch 161, batches 9000|18183, train-acc 0.860, loss 0.366, para-norm 5122.531, grad-norm 5.214, time 7.08s, 
epoch 161, batches 10000|18183, train-acc 0.857, loss 0.375, para-norm 5121.813, grad-norm 3.419, time 7.09s, 
epoch 161, batches 11000|18183, train-acc 0.856, loss 0.376, para-norm 5121.276, grad-norm 4.505, time 7.07s, 
epoch 161, batches 12000|18183, train-acc 0.857, loss 0.375, para-norm 5121.387, grad-norm 5.337, time 7.06s, 
epoch 161, batches 13000|18183, train-acc 0.857, loss 0.376, para-norm 5120.732, grad-norm 5.710, time 7.07s, 
epoch 161, batches 14000|18183, train-acc 0.858, loss 0.370, para-norm 5120.944, grad-norm 21.385, time 7.06s, 
epoch 161, batches 15000|18183, train-acc 0.860, loss 0.371, para-norm 5125.551, grad-norm 8.567, time 7.06s, 
epoch 161, batches 16000|18183, train-acc 0.859, loss 0.370, para-norm 5124.845, grad-norm 6.293, time 7.05s, 
epoch 161, batches 17000|18183, train-acc 0.855, loss 0.379, para-norm 5124.859, grad-norm 5.509, time 7.06s, 
epoch 161, batches 18000|18183, train-acc 0.862, loss 0.365, para-norm 5124.602, grad-norm 25.804, time 7.07s, 
epoch 161, batches 18183|18183, train-acc 0.866, loss 0.351, para-norm 5124.491, grad-norm 5.038, time 1.29s, 
dev-acc 0.859
epoch 162, batches 1000|18183, train-acc 0.859, loss 0.373, para-norm 5123.957, grad-norm 8.366, time 7.14s, 
epoch 162, batches 2000|18183, train-acc 0.862, loss 0.365, para-norm 5123.577, grad-norm 0.072, time 7.15s, 
epoch 162, batches 3000|18183, train-acc 0.863, loss 0.361, para-norm 5122.919, grad-norm 4.376, time 7.05s, 
epoch 162, batches 4000|18183, train-acc 0.856, loss 0.372, para-norm 5122.516, grad-norm 15.319, time 7.08s, 
epoch 162, batches 5000|18183, train-acc 0.859, loss 0.368, para-norm 5122.379, grad-norm 6.640, time 7.06s, 
epoch 162, batches 6000|18183, train-acc 0.857, loss 0.374, para-norm 5121.737, grad-norm 4.834, time 7.08s, 
epoch 162, batches 7000|18183, train-acc 0.857, loss 0.373, para-norm 5121.807, grad-norm 2.618, time 7.08s, 
epoch 162, batches 8000|18183, train-acc 0.858, loss 0.374, para-norm 5121.428, grad-norm 9.090, time 7.06s, 
epoch 162, batches 9000|18183, train-acc 0.855, loss 0.377, para-norm 5120.833, grad-norm 7.852, time 7.33s, 
epoch 162, batches 10000|18183, train-acc 0.859, loss 0.369, para-norm 5120.052, grad-norm 2.346, time 7.40s, 
epoch 162, batches 11000|18183, train-acc 0.857, loss 0.376, para-norm 5119.596, grad-norm 2228.686, time 7.37s, 
epoch 162, batches 12000|18183, train-acc 0.860, loss 0.364, para-norm 5119.740, grad-norm 6.719, time 7.37s, 
epoch 162, batches 13000|18183, train-acc 0.862, loss 0.364, para-norm 5119.484, grad-norm 6.198, time 7.36s, 
epoch 162, batches 14000|18183, train-acc 0.863, loss 0.362, para-norm 5118.990, grad-norm 8.943, time 7.37s, 
epoch 162, batches 15000|18183, train-acc 0.860, loss 0.364, para-norm 5119.164, grad-norm 4.144, time 7.37s, 
epoch 162, batches 16000|18183, train-acc 0.857, loss 0.374, para-norm 5118.969, grad-norm 4.601, time 7.38s, 
epoch 162, batches 17000|18183, train-acc 0.858, loss 0.372, para-norm 5118.369, grad-norm 4.745, time 7.38s, 
epoch 162, batches 18000|18183, train-acc 0.857, loss 0.369, para-norm 5118.011, grad-norm 7.172, time 7.39s, 
epoch 162, batches 18183|18183, train-acc 0.861, loss 0.363, para-norm 5117.803, grad-norm 2.479, time 1.40s, 
dev-acc 0.860
epoch 163, batches 1000|18183, train-acc 0.859, loss 0.369, para-norm 5117.339, grad-norm 1.615, time 7.34s, 
epoch 163, batches 2000|18183, train-acc 0.862, loss 0.365, para-norm 5117.564, grad-norm 3.334, time 7.08s, 
epoch 163, batches 3000|18183, train-acc 0.855, loss 0.379, para-norm 5117.512, grad-norm 7.784, time 7.10s, 
epoch 163, batches 4000|18183, train-acc 0.862, loss 0.364, para-norm 5117.984, grad-norm 1.898, time 7.10s, 
epoch 163, batches 5000|18183, train-acc 0.863, loss 0.361, para-norm 5118.156, grad-norm 10.829, time 7.09s, 
epoch 163, batches 6000|18183, train-acc 0.859, loss 0.370, para-norm 5124.351, grad-norm 3.523, time 7.10s, 
epoch 163, batches 7000|18183, train-acc 0.859, loss 0.365, para-norm 5123.855, grad-norm 2.115, time 7.09s, 
epoch 163, batches 8000|18183, train-acc 0.864, loss 0.361, para-norm 5124.371, grad-norm 10.260, time 7.09s, 
epoch 163, batches 9000|18183, train-acc 0.857, loss 0.372, para-norm 5123.090, grad-norm 5.126, time 7.10s, 
epoch 163, batches 10000|18183, train-acc 0.858, loss 0.374, para-norm 5123.249, grad-norm 5.216, time 7.10s, 
epoch 163, batches 11000|18183, train-acc 0.860, loss 0.370, para-norm 5121.945, grad-norm 4.605, time 7.09s, 
epoch 163, batches 12000|18183, train-acc 0.860, loss 0.368, para-norm 5121.342, grad-norm 4.938, time 7.09s, 
epoch 163, batches 13000|18183, train-acc 0.857, loss 0.377, para-norm 5121.676, grad-norm 10.444, time 7.08s, 
epoch 163, batches 14000|18183, train-acc 0.855, loss 0.376, para-norm 5121.592, grad-norm 7.744, time 7.09s, 
epoch 163, batches 15000|18183, train-acc 0.858, loss 0.373, para-norm 5120.441, grad-norm 9.568, time 7.11s, 
epoch 163, batches 16000|18183, train-acc 0.860, loss 0.369, para-norm 5121.271, grad-norm 8.033, time 7.10s, 
epoch 163, batches 17000|18183, train-acc 0.861, loss 0.370, para-norm 5120.806, grad-norm 2.576, time 7.09s, 
epoch 163, batches 18000|18183, train-acc 0.860, loss 0.364, para-norm 5121.144, grad-norm 4.099, time 7.10s, 
epoch 163, batches 18183|18183, train-acc 0.853, loss 0.398, para-norm 5125.045, grad-norm 4.117, time 1.30s, 
dev-acc 0.860
epoch 164, batches 1000|18183, train-acc 0.862, loss 0.362, para-norm 5126.830, grad-norm 2.863, time 7.16s, 
epoch 164, batches 2000|18183, train-acc 0.860, loss 0.369, para-norm 5126.595, grad-norm 4.904, time 7.10s, 
epoch 164, batches 3000|18183, train-acc 0.861, loss 0.363, para-norm 5125.934, grad-norm 1.519, time 7.09s, 
epoch 164, batches 4000|18183, train-acc 0.860, loss 0.369, para-norm 5126.399, grad-norm 1.039, time 7.09s, 
epoch 164, batches 5000|18183, train-acc 0.861, loss 0.365, para-norm 5126.555, grad-norm 13.778, time 7.10s, 
epoch 164, batches 6000|18183, train-acc 0.861, loss 0.367, para-norm 5126.670, grad-norm 5.875, time 7.08s, 
epoch 164, batches 7000|18183, train-acc 0.859, loss 0.376, para-norm 5126.137, grad-norm 7.616, time 7.09s, 
epoch 164, batches 8000|18183, train-acc 0.859, loss 0.370, para-norm 5125.476, grad-norm 11.919, time 7.09s, 
epoch 164, batches 9000|18183, train-acc 0.862, loss 0.360, para-norm 5125.337, grad-norm 6.930, time 7.09s, 
epoch 164, batches 10000|18183, train-acc 0.862, loss 0.362, para-norm 5125.617, grad-norm 2.460, time 7.10s, 
epoch 164, batches 11000|18183, train-acc 0.859, loss 0.371, para-norm 5125.784, grad-norm 1.355, time 7.09s, 
epoch 164, batches 12000|18183, train-acc 0.858, loss 0.371, para-norm 5132.183, grad-norm 6.518, time 7.09s, 
epoch 164, batches 13000|18183, train-acc 0.857, loss 0.375, para-norm 5132.045, grad-norm 3.016, time 7.09s, 
epoch 164, batches 14000|18183, train-acc 0.856, loss 0.376, para-norm 5131.936, grad-norm 2.297, time 7.10s, 
epoch 164, batches 15000|18183, train-acc 0.858, loss 0.372, para-norm 5131.465, grad-norm 2.980, time 7.08s, 
epoch 164, batches 16000|18183, train-acc 0.856, loss 0.375, para-norm 5130.597, grad-norm 2.693, time 7.10s, 
epoch 164, batches 17000|18183, train-acc 0.857, loss 0.376, para-norm 5130.520, grad-norm 3.569, time 7.16s, 
epoch 164, batches 18000|18183, train-acc 0.857, loss 0.374, para-norm 5129.837, grad-norm 5.834, time 7.08s, 
epoch 164, batches 18183|18183, train-acc 0.859, loss 0.367, para-norm 5129.941, grad-norm 99.492, time 1.29s, 
dev-acc 0.862
epoch 165, batches 1000|18183, train-acc 0.862, loss 0.361, para-norm 5129.748, grad-norm 8.918, time 7.13s, 
epoch 165, batches 2000|18183, train-acc 0.860, loss 0.363, para-norm 5129.388, grad-norm 4.152, time 7.07s, 
epoch 165, batches 3000|18183, train-acc 0.862, loss 0.360, para-norm 5129.738, grad-norm 10.026, time 7.08s, 
epoch 165, batches 4000|18183, train-acc 0.856, loss 0.375, para-norm 5129.102, grad-norm 4.270, time 7.06s, 
epoch 165, batches 5000|18183, train-acc 0.860, loss 0.368, para-norm 5129.052, grad-norm 13.166, time 7.08s, 
epoch 165, batches 6000|18183, train-acc 0.859, loss 0.371, para-norm 5127.973, grad-norm 3.574, time 7.08s, 
epoch 165, batches 7000|18183, train-acc 0.861, loss 0.369, para-norm 5127.758, grad-norm 1.425, time 7.06s, 
epoch 165, batches 8000|18183, train-acc 0.861, loss 0.366, para-norm 5127.957, grad-norm 6.355, time 7.08s, 
epoch 165, batches 9000|18183, train-acc 0.865, loss 0.360, para-norm 5128.206, grad-norm 4.353, time 7.06s, 
epoch 165, batches 10000|18183, train-acc 0.860, loss 0.362, para-norm 5128.135, grad-norm 8.520, time 7.10s, 
epoch 165, batches 11000|18183, train-acc 0.858, loss 0.372, para-norm 5127.634, grad-norm 10.366, time 7.09s, 
epoch 165, batches 12000|18183, train-acc 0.858, loss 0.373, para-norm 5127.578, grad-norm 10.876, time 7.07s, 
epoch 165, batches 13000|18183, train-acc 0.859, loss 0.368, para-norm 5127.068, grad-norm 2.428, time 7.08s, 
epoch 165, batches 14000|18183, train-acc 0.857, loss 0.374, para-norm 5126.943, grad-norm 3.021, time 7.09s, 
epoch 165, batches 15000|18183, train-acc 0.861, loss 0.369, para-norm 5126.632, grad-norm 3.696, time 7.08s, 
epoch 165, batches 16000|18183, train-acc 0.857, loss 0.373, para-norm 5126.343, grad-norm 6.112, time 7.08s, 
epoch 165, batches 17000|18183, train-acc 0.861, loss 0.368, para-norm 5125.524, grad-norm 9.657, time 7.09s, 
epoch 165, batches 18000|18183, train-acc 0.857, loss 0.368, para-norm 5125.293, grad-norm 4.386, time 7.11s, 
epoch 165, batches 18183|18183, train-acc 0.855, loss 0.389, para-norm 5124.934, grad-norm 8.738, time 1.30s, 
dev-acc 0.861
epoch 166, batches 1000|18183, train-acc 0.865, loss 0.356, para-norm 5125.532, grad-norm 6.274, time 7.18s, 
epoch 166, batches 2000|18183, train-acc 0.861, loss 0.364, para-norm 5126.084, grad-norm 6.455, time 7.09s, 
epoch 166, batches 3000|18183, train-acc 0.861, loss 0.367, para-norm 5125.860, grad-norm 10.476, time 7.07s, 
epoch 166, batches 4000|18183, train-acc 0.863, loss 0.362, para-norm 5126.938, grad-norm 5.473, time 7.10s, 
epoch 166, batches 5000|18183, train-acc 0.860, loss 0.365, para-norm 5126.036, grad-norm 7.425, time 7.09s, 
epoch 166, batches 6000|18183, train-acc 0.858, loss 0.373, para-norm 5126.143, grad-norm 4.547, time 7.16s, 
epoch 166, batches 7000|18183, train-acc 0.860, loss 0.371, para-norm 5125.794, grad-norm 2.953, time 7.25s, 
epoch 166, batches 8000|18183, train-acc 0.861, loss 0.364, para-norm 5125.125, grad-norm 5.917, time 7.54s, 
epoch 166, batches 9000|18183, train-acc 0.858, loss 0.372, para-norm 5124.978, grad-norm 1.247, time 7.27s, 
epoch 166, batches 10000|18183, train-acc 0.859, loss 0.368, para-norm 5124.400, grad-norm 4.043, time 7.24s, 
epoch 166, batches 11000|18183, train-acc 0.860, loss 0.367, para-norm 5123.526, grad-norm 4.375, time 7.23s, 
epoch 166, batches 12000|18183, train-acc 0.858, loss 0.369, para-norm 5125.178, grad-norm 14.156, time 7.24s, 
epoch 166, batches 13000|18183, train-acc 0.859, loss 0.370, para-norm 5124.670, grad-norm 3.469, time 7.24s, 
epoch 166, batches 14000|18183, train-acc 0.858, loss 0.370, para-norm 5124.273, grad-norm 3.389, time 7.25s, 
epoch 166, batches 15000|18183, train-acc 0.856, loss 0.374, para-norm 5123.655, grad-norm 3.471, time 7.23s, 
epoch 166, batches 16000|18183, train-acc 0.859, loss 0.373, para-norm 5122.965, grad-norm 9.017, time 7.23s, 
epoch 166, batches 17000|18183, train-acc 0.859, loss 0.370, para-norm 5122.911, grad-norm 9.660, time 7.24s, 
epoch 166, batches 18000|18183, train-acc 0.858, loss 0.370, para-norm 5122.656, grad-norm 2.102, time 7.24s, 
epoch 166, batches 18183|18183, train-acc 0.865, loss 0.360, para-norm 5122.548, grad-norm 3.786, time 1.48s, 
dev-acc 0.860
epoch 167, batches 1000|18183, train-acc 0.859, loss 0.364, para-norm 5123.055, grad-norm 3.318, time 7.25s, 
epoch 167, batches 2000|18183, train-acc 0.859, loss 0.368, para-norm 5123.232, grad-norm 3.724, time 7.23s, 
epoch 167, batches 3000|18183, train-acc 0.858, loss 0.373, para-norm 5122.809, grad-norm 5.007, time 7.24s, 
epoch 167, batches 4000|18183, train-acc 0.859, loss 0.373, para-norm 5122.641, grad-norm 4.624, time 7.22s, 
epoch 167, batches 5000|18183, train-acc 0.858, loss 0.369, para-norm 5122.725, grad-norm 0.016, time 7.21s, 
epoch 167, batches 6000|18183, train-acc 0.862, loss 0.364, para-norm 5122.717, grad-norm 3.760, time 7.22s, 
epoch 167, batches 7000|18183, train-acc 0.862, loss 0.363, para-norm 5122.287, grad-norm 1.517, time 7.21s, 
epoch 167, batches 8000|18183, train-acc 0.863, loss 0.361, para-norm 5122.462, grad-norm 2.858, time 7.21s, 
epoch 167, batches 9000|18183, train-acc 0.859, loss 0.367, para-norm 5122.566, grad-norm 8.648, time 7.22s, 
epoch 167, batches 10000|18183, train-acc 0.864, loss 0.362, para-norm 5122.478, grad-norm 3.169, time 7.22s, 
epoch 167, batches 11000|18183, train-acc 0.860, loss 0.366, para-norm 5122.116, grad-norm 4.710, time 7.22s, 
epoch 167, batches 12000|18183, train-acc 0.857, loss 0.370, para-norm 5122.097, grad-norm 5.228, time 7.22s, 
epoch 167, batches 13000|18183, train-acc 0.859, loss 0.369, para-norm 5121.161, grad-norm 4.044, time 7.22s, 
epoch 167, batches 14000|18183, train-acc 0.859, loss 0.369, para-norm 5120.809, grad-norm 6.897, time 7.04s, 
epoch 167, batches 15000|18183, train-acc 0.861, loss 0.365, para-norm 5121.175, grad-norm 3.541, time 6.89s, 
epoch 167, batches 16000|18183, train-acc 0.857, loss 0.370, para-norm 5121.211, grad-norm 11.920, time 6.91s, 
epoch 167, batches 17000|18183, train-acc 0.859, loss 0.371, para-norm 5120.679, grad-norm 3.058, time 6.90s, 
epoch 167, batches 18000|18183, train-acc 0.859, loss 0.374, para-norm 5120.360, grad-norm 3.874, time 6.91s, 
epoch 167, batches 18183|18183, train-acc 0.863, loss 0.358, para-norm 5120.506, grad-norm 6.376, time 1.26s, 
dev-acc 0.858
epoch 168, batches 1000|18183, train-acc 0.861, loss 0.361, para-norm 5120.549, grad-norm 5.307, time 6.89s, 
epoch 168, batches 2000|18183, train-acc 0.859, loss 0.373, para-norm 5120.115, grad-norm 5.539, time 6.91s, 
epoch 168, batches 3000|18183, train-acc 0.861, loss 0.364, para-norm 5121.012, grad-norm 5.982, time 6.90s, 
epoch 168, batches 4000|18183, train-acc 0.860, loss 0.367, para-norm 5120.372, grad-norm 3.044, time 6.93s, 
epoch 168, batches 5000|18183, train-acc 0.859, loss 0.369, para-norm 5120.775, grad-norm 4.809, time 6.91s, 
epoch 168, batches 6000|18183, train-acc 0.860, loss 0.371, para-norm 5120.836, grad-norm 6.026, time 6.91s, 
epoch 168, batches 7000|18183, train-acc 0.860, loss 0.368, para-norm 5120.668, grad-norm 3.113, time 7.14s, 
epoch 168, batches 8000|18183, train-acc 0.861, loss 0.367, para-norm 5120.064, grad-norm 0.712, time 7.12s, 
epoch 168, batches 9000|18183, train-acc 0.857, loss 0.372, para-norm 5119.578, grad-norm 1237.943, time 7.15s, 
epoch 168, batches 10000|18183, train-acc 0.861, loss 0.362, para-norm 5119.156, grad-norm 14.561, time 7.04s, 
epoch 168, batches 11000|18183, train-acc 0.860, loss 0.368, para-norm 5119.292, grad-norm 4.561, time 7.05s, 
epoch 168, batches 12000|18183, train-acc 0.861, loss 0.363, para-norm 5120.058, grad-norm 3.425, time 7.09s, 
epoch 168, batches 13000|18183, train-acc 0.856, loss 0.377, para-norm 5119.557, grad-norm 7.240, time 7.07s, 
epoch 168, batches 14000|18183, train-acc 0.863, loss 0.361, para-norm 5119.675, grad-norm 6.001, time 7.13s, 
epoch 168, batches 15000|18183, train-acc 0.859, loss 0.365, para-norm 5119.796, grad-norm 1.446, time 7.06s, 
epoch 168, batches 16000|18183, train-acc 0.861, loss 0.365, para-norm 5119.459, grad-norm 10.900, time 7.05s, 
epoch 168, batches 17000|18183, train-acc 0.860, loss 0.369, para-norm 5118.708, grad-norm 3.162, time 7.06s, 
epoch 168, batches 18000|18183, train-acc 0.859, loss 0.371, para-norm 5118.987, grad-norm 6.997, time 7.14s, 
epoch 168, batches 18183|18183, train-acc 0.861, loss 0.365, para-norm 5118.901, grad-norm 3.792, time 1.30s, 
dev-acc 0.858
epoch 169, batches 1000|18183, train-acc 0.863, loss 0.358, para-norm 5119.613, grad-norm 4.338, time 7.16s, 
epoch 169, batches 2000|18183, train-acc 0.860, loss 0.367, para-norm 5120.421, grad-norm 5.828, time 7.48s, 
epoch 169, batches 3000|18183, train-acc 0.863, loss 0.362, para-norm 5120.119, grad-norm 2.568, time 7.34s, 
epoch 169, batches 4000|18183, train-acc 0.861, loss 0.361, para-norm 5120.176, grad-norm 1.328, time 7.24s, 
epoch 169, batches 5000|18183, train-acc 0.862, loss 0.360, para-norm 5120.657, grad-norm 4.740, time 7.23s, 
epoch 169, batches 6000|18183, train-acc 0.860, loss 0.371, para-norm 5119.749, grad-norm 1.514, time 7.24s, 
epoch 169, batches 7000|18183, train-acc 0.861, loss 0.368, para-norm 5123.507, grad-norm 6.270, time 7.36s, 
epoch 169, batches 8000|18183, train-acc 0.862, loss 0.367, para-norm 5122.371, grad-norm 5.009, time 7.23s, 
epoch 169, batches 9000|18183, train-acc 0.858, loss 0.372, para-norm 5122.024, grad-norm 7.455, time 7.26s, 
epoch 169, batches 10000|18183, train-acc 0.860, loss 0.359, para-norm 5121.833, grad-norm 6.812, time 7.22s, 
epoch 169, batches 11000|18183, train-acc 0.861, loss 0.366, para-norm 5121.305, grad-norm 1.741, time 7.24s, 
epoch 169, batches 12000|18183, train-acc 0.862, loss 0.367, para-norm 5120.943, grad-norm 6.969, time 7.25s, 
epoch 169, batches 13000|18183, train-acc 0.861, loss 0.367, para-norm 5121.001, grad-norm 3.414, time 7.24s, 
epoch 169, batches 14000|18183, train-acc 0.860, loss 0.372, para-norm 5120.229, grad-norm 5.774, time 7.23s, 
epoch 169, batches 15000|18183, train-acc 0.860, loss 0.368, para-norm 5119.344, grad-norm 21.029, time 7.19s, 
epoch 169, batches 16000|18183, train-acc 0.856, loss 0.370, para-norm 5118.503, grad-norm 13.553, time 7.13s, 
epoch 169, batches 17000|18183, train-acc 0.863, loss 0.364, para-norm 5118.537, grad-norm 2.117, time 7.16s, 
epoch 169, batches 18000|18183, train-acc 0.858, loss 0.369, para-norm 5118.371, grad-norm 5.340, time 7.07s, 
epoch 169, batches 18183|18183, train-acc 0.855, loss 0.376, para-norm 5118.099, grad-norm 5.484, time 1.29s, 
dev-acc 0.858
epoch 170, batches 1000|18183, train-acc 0.863, loss 0.357, para-norm 5117.978, grad-norm 7.399, time 7.12s, 
epoch 170, batches 2000|18183, train-acc 0.865, loss 0.360, para-norm 5118.366, grad-norm 6.536, time 7.06s, 
epoch 170, batches 3000|18183, train-acc 0.862, loss 0.363, para-norm 5118.180, grad-norm 6.331, time 7.06s, 
epoch 170, batches 4000|18183, train-acc 0.860, loss 0.369, para-norm 5118.710, grad-norm 3.303, time 7.11s, 
epoch 170, batches 5000|18183, train-acc 0.860, loss 0.366, para-norm 5118.546, grad-norm 6.324, time 7.07s, 
epoch 170, batches 6000|18183, train-acc 0.860, loss 0.370, para-norm 5118.314, grad-norm 3.975, time 7.07s, 
epoch 170, batches 7000|18183, train-acc 0.860, loss 0.370, para-norm 5118.153, grad-norm 7.191, time 7.17s, 
epoch 170, batches 8000|18183, train-acc 0.859, loss 0.366, para-norm 5118.327, grad-norm 11.520, time 7.16s, 
epoch 170, batches 9000|18183, train-acc 0.861, loss 0.366, para-norm 5119.424, grad-norm 4.414, time 7.22s, 
epoch 170, batches 10000|18183, train-acc 0.861, loss 0.369, para-norm 5118.134, grad-norm 2.193, time 7.25s, 
epoch 170, batches 11000|18183, train-acc 0.858, loss 0.370, para-norm 5117.991, grad-norm 11.209, time 7.23s, 
epoch 170, batches 12000|18183, train-acc 0.864, loss 0.356, para-norm 5117.544, grad-norm 4.509, time 7.22s, 
epoch 170, batches 13000|18183, train-acc 0.858, loss 0.371, para-norm 5117.063, grad-norm 6.959, time 7.24s, 
epoch 170, batches 14000|18183, train-acc 0.858, loss 0.371, para-norm 5117.396, grad-norm 4.627, time 7.24s, 
epoch 170, batches 15000|18183, train-acc 0.857, loss 0.372, para-norm 5117.580, grad-norm 3.260, time 7.36s, 
epoch 170, batches 16000|18183, train-acc 0.862, loss 0.368, para-norm 5117.412, grad-norm 3.790, time 7.10s, 
epoch 170, batches 17000|18183, train-acc 0.859, loss 0.369, para-norm 5116.506, grad-norm 5.119, time 7.07s, 
epoch 170, batches 18000|18183, train-acc 0.863, loss 0.361, para-norm 5117.111, grad-norm 4.616, time 7.07s, 
epoch 170, batches 18183|18183, train-acc 0.862, loss 0.359, para-norm 5116.874, grad-norm 3.512, time 1.29s, 
dev-acc 0.862
epoch 171, batches 1000|18183, train-acc 0.864, loss 0.360, para-norm 5116.619, grad-norm 206.418, time 7.15s, 
epoch 171, batches 2000|18183, train-acc 0.861, loss 0.370, para-norm 5116.482, grad-norm 5.925, time 7.08s, 
epoch 171, batches 3000|18183, train-acc 0.862, loss 0.366, para-norm 5116.540, grad-norm 7.439, time 7.04s, 
epoch 171, batches 4000|18183, train-acc 0.863, loss 0.361, para-norm 5116.128, grad-norm 4.406, time 7.05s, 
epoch 171, batches 5000|18183, train-acc 0.860, loss 0.366, para-norm 5116.288, grad-norm 4.838, time 7.07s, 
epoch 171, batches 6000|18183, train-acc 0.864, loss 0.358, para-norm 5116.491, grad-norm 4.227, time 7.06s, 
epoch 171, batches 7000|18183, train-acc 0.857, loss 0.372, para-norm 5116.432, grad-norm 11.620, time 7.09s, 
epoch 171, batches 8000|18183, train-acc 0.862, loss 0.367, para-norm 5116.889, grad-norm 3.648, time 7.08s, 
epoch 171, batches 9000|18183, train-acc 0.859, loss 0.369, para-norm 5116.438, grad-norm 14.792, time 7.08s, 
epoch 171, batches 10000|18183, train-acc 0.860, loss 0.366, para-norm 5116.455, grad-norm 3.249, time 7.06s, 
epoch 171, batches 11000|18183, train-acc 0.860, loss 0.366, para-norm 5116.491, grad-norm 2.755, time 7.18s, 
epoch 171, batches 12000|18183, train-acc 0.860, loss 0.363, para-norm 5116.448, grad-norm 1.454, time 7.08s, 
epoch 171, batches 13000|18183, train-acc 0.860, loss 0.370, para-norm 5116.366, grad-norm 2.730, time 7.07s, 
epoch 171, batches 14000|18183, train-acc 0.861, loss 0.368, para-norm 5116.817, grad-norm 8.520, time 7.06s, 
epoch 171, batches 15000|18183, train-acc 0.862, loss 0.364, para-norm 5116.806, grad-norm 7.076, time 7.06s, 
epoch 171, batches 16000|18183, train-acc 0.858, loss 0.367, para-norm 5116.080, grad-norm 3.201, time 7.07s, 
epoch 171, batches 17000|18183, train-acc 0.861, loss 0.366, para-norm 5116.445, grad-norm 9.802, time 7.06s, 
epoch 171, batches 18000|18183, train-acc 0.860, loss 0.366, para-norm 5116.431, grad-norm 3.966, time 7.06s, 
epoch 171, batches 18183|18183, train-acc 0.859, loss 0.363, para-norm 5116.478, grad-norm 4.185, time 1.29s, 
dev-acc 0.860
epoch 172, batches 1000|18183, train-acc 0.859, loss 0.370, para-norm 5116.735, grad-norm 5.165, time 7.40s, 
epoch 172, batches 2000|18183, train-acc 0.863, loss 0.360, para-norm 5116.473, grad-norm 6.616, time 7.06s, 
epoch 172, batches 3000|18183, train-acc 0.857, loss 0.369, para-norm 5116.441, grad-norm 3.866, time 7.08s, 
epoch 172, batches 4000|18183, train-acc 0.860, loss 0.366, para-norm 5116.169, grad-norm 10.683, time 7.06s, 
epoch 172, batches 5000|18183, train-acc 0.857, loss 0.373, para-norm 5115.946, grad-norm 9.234, time 7.08s, 
epoch 172, batches 6000|18183, train-acc 0.857, loss 0.369, para-norm 5115.614, grad-norm 5.079, time 7.07s, 
epoch 172, batches 7000|18183, train-acc 0.865, loss 0.358, para-norm 5116.117, grad-norm 2.074, time 7.13s, 
epoch 172, batches 8000|18183, train-acc 0.865, loss 0.359, para-norm 5115.625, grad-norm 4.055, time 7.06s, 
epoch 172, batches 9000|18183, train-acc 0.859, loss 0.370, para-norm 5115.908, grad-norm 290.960, time 7.08s, 
epoch 172, batches 10000|18183, train-acc 0.862, loss 0.361, para-norm 5116.607, grad-norm 8.857, time 7.05s, 
epoch 172, batches 11000|18183, train-acc 0.865, loss 0.353, para-norm 5116.730, grad-norm 4.150, time 7.07s, 
epoch 172, batches 12000|18183, train-acc 0.861, loss 0.366, para-norm 5116.015, grad-norm 5.420, time 7.08s, 
epoch 172, batches 13000|18183, train-acc 0.856, loss 0.372, para-norm 5115.867, grad-norm 4.542, time 7.08s, 
epoch 172, batches 14000|18183, train-acc 0.860, loss 0.368, para-norm 5115.543, grad-norm 2.941, time 7.06s, 
epoch 172, batches 15000|18183, train-acc 0.859, loss 0.370, para-norm 5114.843, grad-norm 5.137, time 7.08s, 
epoch 172, batches 16000|18183, train-acc 0.859, loss 0.368, para-norm 5114.610, grad-norm 5.006, time 7.08s, 
epoch 172, batches 17000|18183, train-acc 0.860, loss 0.366, para-norm 5115.390, grad-norm 9.597, time 7.08s, 
epoch 172, batches 18000|18183, train-acc 0.859, loss 0.371, para-norm 5114.580, grad-norm 10.439, time 7.08s, 
epoch 172, batches 18183|18183, train-acc 0.859, loss 0.372, para-norm 5114.453, grad-norm 4.018, time 1.30s, 
dev-acc 0.859
epoch 173, batches 1000|18183, train-acc 0.855, loss 0.375, para-norm 5114.575, grad-norm 2.602, time 7.21s, 
epoch 173, batches 2000|18183, train-acc 0.862, loss 0.362, para-norm 5114.682, grad-norm 4.075, time 7.06s, 
epoch 173, batches 3000|18183, train-acc 0.863, loss 0.356, para-norm 5115.296, grad-norm 11.252, time 7.14s, 
epoch 173, batches 4000|18183, train-acc 0.860, loss 0.367, para-norm 5114.743, grad-norm 6.250, time 7.06s, 
epoch 173, batches 5000|18183, train-acc 0.863, loss 0.363, para-norm 5114.141, grad-norm 5.106, time 7.10s, 
epoch 173, batches 6000|18183, train-acc 0.858, loss 0.367, para-norm 5114.271, grad-norm 7.373, time 7.15s, 
epoch 173, batches 7000|18183, train-acc 0.864, loss 0.360, para-norm 5114.598, grad-norm 7.338, time 7.10s, 
epoch 173, batches 8000|18183, train-acc 0.863, loss 0.360, para-norm 5114.634, grad-norm 4.966, time 7.06s, 
epoch 173, batches 9000|18183, train-acc 0.860, loss 0.369, para-norm 5114.310, grad-norm 5.258, time 7.11s, 
epoch 173, batches 10000|18183, train-acc 0.859, loss 0.364, para-norm 5114.685, grad-norm 4.327, time 7.13s, 
epoch 173, batches 11000|18183, train-acc 0.859, loss 0.372, para-norm 5114.543, grad-norm 7.120, time 7.09s, 
epoch 173, batches 12000|18183, train-acc 0.866, loss 0.356, para-norm 5114.483, grad-norm 3.207, time 7.09s, 
epoch 173, batches 13000|18183, train-acc 0.865, loss 0.363, para-norm 5114.274, grad-norm 11.307, time 7.10s, 
epoch 173, batches 14000|18183, train-acc 0.861, loss 0.368, para-norm 5113.863, grad-norm 5.991, time 7.09s, 
epoch 173, batches 15000|18183, train-acc 0.862, loss 0.364, para-norm 5114.343, grad-norm 4.214, time 7.06s, 
epoch 173, batches 16000|18183, train-acc 0.859, loss 0.369, para-norm 5116.341, grad-norm 4.576, time 7.13s, 
epoch 173, batches 17000|18183, train-acc 0.860, loss 0.365, para-norm 5115.903, grad-norm 4.793, time 6.99s, 
epoch 173, batches 18000|18183, train-acc 0.858, loss 0.367, para-norm 5121.531, grad-norm 9.432, time 7.08s, 
epoch 173, batches 18183|18183, train-acc 0.859, loss 0.367, para-norm 5122.068, grad-norm 10.256, time 1.30s, 
dev-acc 0.860
epoch 174, batches 1000|18183, train-acc 0.861, loss 0.361, para-norm 5123.941, grad-norm 5.075, time 7.25s, 
epoch 174, batches 2000|18183, train-acc 0.857, loss 0.374, para-norm 5122.951, grad-norm 2.692, time 7.09s, 
epoch 174, batches 3000|18183, train-acc 0.858, loss 0.374, para-norm 5122.250, grad-norm 2.933, time 7.09s, 
epoch 174, batches 4000|18183, train-acc 0.863, loss 0.358, para-norm 5121.958, grad-norm 9.003, time 7.06s, 
epoch 174, batches 5000|18183, train-acc 0.860, loss 0.365, para-norm 5122.290, grad-norm 7.095, time 7.08s, 
epoch 174, batches 6000|18183, train-acc 0.863, loss 0.362, para-norm 5122.316, grad-norm 7.132, time 7.08s, 
epoch 174, batches 7000|18183, train-acc 0.864, loss 0.360, para-norm 5121.618, grad-norm 7.801, time 7.06s, 
epoch 174, batches 8000|18183, train-acc 0.865, loss 0.359, para-norm 5120.930, grad-norm 1.984, time 7.14s, 
epoch 174, batches 9000|18183, train-acc 0.863, loss 0.361, para-norm 5120.717, grad-norm 5.022, time 7.15s, 
epoch 174, batches 10000|18183, train-acc 0.859, loss 0.369, para-norm 5119.634, grad-norm 2.175, time 7.09s, 
epoch 174, batches 11000|18183, train-acc 0.864, loss 0.362, para-norm 5119.746, grad-norm 9.401, time 7.07s, 
epoch 174, batches 12000|18183, train-acc 0.865, loss 0.357, para-norm 5120.133, grad-norm 2.669, time 7.07s, 
epoch 174, batches 13000|18183, train-acc 0.862, loss 0.365, para-norm 5120.393, grad-norm 3.702, time 7.07s, 
epoch 174, batches 14000|18183, train-acc 0.861, loss 0.366, para-norm 5120.270, grad-norm 7.189, time 7.06s, 
epoch 174, batches 15000|18183, train-acc 0.862, loss 0.367, para-norm 5120.255, grad-norm 17.703, time 7.07s, 
epoch 174, batches 16000|18183, train-acc 0.859, loss 0.365, para-norm 5120.155, grad-norm 4.909, time 7.07s, 
epoch 174, batches 17000|18183, train-acc 0.861, loss 0.370, para-norm 5121.259, grad-norm 10.993, time 7.09s, 
epoch 174, batches 18000|18183, train-acc 0.862, loss 0.364, para-norm 5120.913, grad-norm 12.051, time 7.08s, 
epoch 174, batches 18183|18183, train-acc 0.860, loss 0.365, para-norm 5121.128, grad-norm 4.684, time 1.29s, 
dev-acc 0.861
epoch 175, batches 1000|18183, train-acc 0.866, loss 0.358, para-norm 5121.130, grad-norm 5.573, time 7.11s, 
epoch 175, batches 2000|18183, train-acc 0.865, loss 0.358, para-norm 5120.479, grad-norm 5.868, time 7.16s, 
epoch 175, batches 3000|18183, train-acc 0.860, loss 0.365, para-norm 5119.736, grad-norm 1.774, time 7.09s, 
epoch 175, batches 4000|18183, train-acc 0.861, loss 0.361, para-norm 5120.018, grad-norm 1.570, time 7.09s, 
epoch 175, batches 5000|18183, train-acc 0.862, loss 0.363, para-norm 5119.966, grad-norm 7.571, time 7.07s, 
epoch 175, batches 6000|18183, train-acc 0.859, loss 0.364, para-norm 5118.608, grad-norm 3.995, time 7.10s, 
epoch 175, batches 7000|18183, train-acc 0.861, loss 0.365, para-norm 5118.669, grad-norm 4.539, time 7.09s, 
epoch 175, batches 8000|18183, train-acc 0.862, loss 0.366, para-norm 5118.357, grad-norm 7.106, time 7.07s, 
epoch 175, batches 9000|18183, train-acc 0.860, loss 0.367, para-norm 5118.395, grad-norm 4.095, time 7.08s, 
epoch 175, batches 10000|18183, train-acc 0.860, loss 0.369, para-norm 5118.537, grad-norm 3.237, time 7.07s, 
epoch 175, batches 11000|18183, train-acc 0.863, loss 0.362, para-norm 5118.041, grad-norm 12.033, time 7.06s, 
epoch 175, batches 12000|18183, train-acc 0.861, loss 0.363, para-norm 5117.509, grad-norm 6.748, time 7.15s, 
epoch 175, batches 13000|18183, train-acc 0.860, loss 0.370, para-norm 5117.616, grad-norm 6.833, time 7.08s, 
epoch 175, batches 14000|18183, train-acc 0.860, loss 0.365, para-norm 5117.698, grad-norm 7.522, time 7.08s, 
epoch 175, batches 15000|18183, train-acc 0.861, loss 0.366, para-norm 5117.278, grad-norm 7.387, time 7.05s, 
epoch 175, batches 16000|18183, train-acc 0.860, loss 0.366, para-norm 5117.451, grad-norm 4.460, time 7.24s, 
epoch 175, batches 17000|18183, train-acc 0.858, loss 0.371, para-norm 5117.431, grad-norm 4.766, time 7.10s, 
epoch 175, batches 18000|18183, train-acc 0.863, loss 0.361, para-norm 5117.642, grad-norm 5.641, time 7.09s, 
epoch 175, batches 18183|18183, train-acc 0.860, loss 0.369, para-norm 5117.655, grad-norm 5.622, time 1.30s, 
dev-acc 0.860
epoch 176, batches 1000|18183, train-acc 0.864, loss 0.362, para-norm 5118.172, grad-norm 4.836, time 7.17s, 
epoch 176, batches 2000|18183, train-acc 0.860, loss 0.369, para-norm 5117.611, grad-norm 2.097, time 7.07s, 
epoch 176, batches 3000|18183, train-acc 0.861, loss 0.362, para-norm 5123.554, grad-norm 3.219, time 7.06s, 
epoch 176, batches 4000|18183, train-acc 0.864, loss 0.358, para-norm 5123.468, grad-norm 3.825, time 7.07s, 
epoch 176, batches 5000|18183, train-acc 0.861, loss 0.360, para-norm 5123.655, grad-norm 1.920, time 7.13s, 
epoch 176, batches 6000|18183, train-acc 0.862, loss 0.360, para-norm 5122.741, grad-norm 3.395, time 7.08s, 
epoch 176, batches 7000|18183, train-acc 0.857, loss 0.367, para-norm 5122.961, grad-norm 9.178, time 7.08s, 
epoch 176, batches 8000|18183, train-acc 0.860, loss 0.366, para-norm 5122.184, grad-norm 9.119, time 7.09s, 
epoch 176, batches 9000|18183, train-acc 0.861, loss 0.368, para-norm 5121.983, grad-norm 3.731, time 7.08s, 
epoch 176, batches 10000|18183, train-acc 0.861, loss 0.362, para-norm 5122.670, grad-norm 4.227, time 7.08s, 
epoch 176, batches 11000|18183, train-acc 0.861, loss 0.366, para-norm 5123.139, grad-norm 2.877, time 7.09s, 
epoch 176, batches 12000|18183, train-acc 0.863, loss 0.364, para-norm 5123.153, grad-norm 21.383, time 7.08s, 
epoch 176, batches 13000|18183, train-acc 0.861, loss 0.365, para-norm 5122.900, grad-norm 6.849, time 7.10s, 
epoch 176, batches 14000|18183, train-acc 0.857, loss 0.375, para-norm 5122.662, grad-norm 2.643, time 7.10s, 
epoch 176, batches 15000|18183, train-acc 0.860, loss 0.364, para-norm 5122.295, grad-norm 4.816, time 7.09s, 
epoch 176, batches 16000|18183, train-acc 0.864, loss 0.359, para-norm 5121.533, grad-norm 2.462, time 7.08s, 
epoch 176, batches 17000|18183, train-acc 0.861, loss 0.364, para-norm 5120.593, grad-norm 5.296, time 7.09s, 
epoch 176, batches 18000|18183, train-acc 0.862, loss 0.360, para-norm 5120.715, grad-norm 5.085, time 7.09s, 
epoch 176, batches 18183|18183, train-acc 0.866, loss 0.348, para-norm 5120.762, grad-norm 9.014, time 1.29s, 
dev-acc 0.861
epoch 177, batches 1000|18183, train-acc 0.863, loss 0.356, para-norm 5120.499, grad-norm 0.682, time 7.16s, 
epoch 177, batches 2000|18183, train-acc 0.862, loss 0.363, para-norm 5120.727, grad-norm 2.439, time 7.08s, 
epoch 177, batches 3000|18183, train-acc 0.861, loss 0.362, para-norm 5121.234, grad-norm 3.973, time 7.09s, 
epoch 177, batches 4000|18183, train-acc 0.865, loss 0.354, para-norm 5120.561, grad-norm 6.430, time 7.35s, 
epoch 177, batches 5000|18183, train-acc 0.860, loss 0.366, para-norm 5120.522, grad-norm 4.840, time 7.30s, 
epoch 177, batches 6000|18183, train-acc 0.865, loss 0.357, para-norm 5119.826, grad-norm 4.693, time 7.27s, 
epoch 177, batches 7000|18183, train-acc 0.857, loss 0.375, para-norm 5119.536, grad-norm 5.686, time 7.30s, 
epoch 177, batches 8000|18183, train-acc 0.864, loss 0.365, para-norm 5119.024, grad-norm 7.333, time 7.29s, 
epoch 177, batches 9000|18183, train-acc 0.862, loss 0.365, para-norm 5118.175, grad-norm 1.513, time 7.29s, 
epoch 177, batches 10000|18183, train-acc 0.861, loss 0.364, para-norm 5118.023, grad-norm 6.906, time 7.30s, 
epoch 177, batches 11000|18183, train-acc 0.861, loss 0.362, para-norm 5117.812, grad-norm 3.769, time 7.30s, 
epoch 177, batches 12000|18183, train-acc 0.861, loss 0.366, para-norm 5116.896, grad-norm 5.767, time 7.38s, 
epoch 177, batches 13000|18183, train-acc 0.861, loss 0.366, para-norm 5121.821, grad-norm 3.145, time 7.29s, 
epoch 177, batches 14000|18183, train-acc 0.860, loss 0.366, para-norm 5121.679, grad-norm 6.955, time 7.35s, 
epoch 177, batches 15000|18183, train-acc 0.859, loss 0.371, para-norm 5121.997, grad-norm 3.467, time 7.36s, 
epoch 177, batches 16000|18183, train-acc 0.858, loss 0.368, para-norm 5121.701, grad-norm 4.078, time 7.32s, 
epoch 177, batches 17000|18183, train-acc 0.860, loss 0.365, para-norm 5120.897, grad-norm 336.850, time 7.27s, 
epoch 177, batches 18000|18183, train-acc 0.864, loss 0.361, para-norm 5120.864, grad-norm 3.417, time 7.25s, 
epoch 177, batches 18183|18183, train-acc 0.865, loss 0.371, para-norm 5120.895, grad-norm 7.940, time 1.32s, 
dev-acc 0.861
epoch 178, batches 1000|18183, train-acc 0.863, loss 0.357, para-norm 5121.474, grad-norm 2.880, time 8.08s, 
epoch 178, batches 2000|18183, train-acc 0.864, loss 0.361, para-norm 5121.044, grad-norm 6.848, time 8.02s, 
epoch 178, batches 3000|18183, train-acc 0.863, loss 0.364, para-norm 5121.447, grad-norm 4.887, time 8.07s, 
epoch 178, batches 4000|18183, train-acc 0.860, loss 0.366, para-norm 5121.115, grad-norm 5.774, time 7.87s, 
epoch 178, batches 5000|18183, train-acc 0.864, loss 0.361, para-norm 5120.239, grad-norm 4.989, time 7.27s, 
epoch 178, batches 6000|18183, train-acc 0.861, loss 0.359, para-norm 5120.286, grad-norm 2.267, time 7.26s, 
epoch 178, batches 7000|18183, train-acc 0.863, loss 0.356, para-norm 5119.835, grad-norm 5.041, time 7.26s, 
epoch 178, batches 8000|18183, train-acc 0.856, loss 0.377, para-norm 5119.473, grad-norm 0.441, time 7.28s, 
epoch 178, batches 9000|18183, train-acc 0.863, loss 0.362, para-norm 5119.183, grad-norm 7.097, time 7.26s, 
epoch 178, batches 10000|18183, train-acc 0.857, loss 0.368, para-norm 5119.569, grad-norm 5.492, time 7.27s, 
epoch 178, batches 11000|18183, train-acc 0.861, loss 0.362, para-norm 5119.295, grad-norm 13.676, time 7.27s, 
epoch 178, batches 12000|18183, train-acc 0.864, loss 0.358, para-norm 5118.925, grad-norm 15.298, time 7.26s, 
epoch 178, batches 13000|18183, train-acc 0.860, loss 0.367, para-norm 5118.508, grad-norm 8.104, time 7.27s, 
epoch 178, batches 14000|18183, train-acc 0.860, loss 0.368, para-norm 5118.405, grad-norm 7.188, time 7.25s, 
epoch 178, batches 15000|18183, train-acc 0.857, loss 0.374, para-norm 5118.101, grad-norm 4.743, time 7.27s, 
epoch 178, batches 16000|18183, train-acc 0.864, loss 0.363, para-norm 5117.950, grad-norm 8.861, time 7.25s, 
epoch 178, batches 17000|18183, train-acc 0.864, loss 0.361, para-norm 5117.455, grad-norm 3.880, time 7.25s, 
epoch 178, batches 18000|18183, train-acc 0.860, loss 0.365, para-norm 5117.438, grad-norm 7.942, time 7.26s, 
epoch 178, batches 18183|18183, train-acc 0.858, loss 0.372, para-norm 5117.319, grad-norm 17.068, time 1.33s, 
dev-acc 0.859
epoch 179, batches 1000|18183, train-acc 0.864, loss 0.361, para-norm 5117.608, grad-norm 4.282, time 8.03s, 
epoch 179, batches 2000|18183, train-acc 0.863, loss 0.358, para-norm 5117.384, grad-norm 6.089, time 7.27s, 
epoch 179, batches 3000|18183, train-acc 0.863, loss 0.358, para-norm 5117.611, grad-norm 6.697, time 7.27s, 
epoch 179, batches 4000|18183, train-acc 0.864, loss 0.362, para-norm 5117.019, grad-norm 6.674, time 7.26s, 
epoch 179, batches 5000|18183, train-acc 0.865, loss 0.357, para-norm 5117.318, grad-norm 3.377, time 7.50s, 
epoch 179, batches 6000|18183, train-acc 0.864, loss 0.362, para-norm 5117.446, grad-norm 1.587, time 7.93s, 
epoch 179, batches 7000|18183, train-acc 0.862, loss 0.359, para-norm 5117.225, grad-norm 3.294, time 7.25s, 
epoch 179, batches 8000|18183, train-acc 0.863, loss 0.362, para-norm 5117.241, grad-norm 7.264, time 7.31s, 
epoch 179, batches 9000|18183, train-acc 0.863, loss 0.364, para-norm 5116.985, grad-norm 8.172, time 7.58s, 
epoch 179, batches 10000|18183, train-acc 0.860, loss 0.369, para-norm 5116.192, grad-norm 10.312, time 7.37s, 
epoch 179, batches 11000|18183, train-acc 0.862, loss 0.364, para-norm 5116.320, grad-norm 9.629, time 7.25s, 
epoch 179, batches 12000|18183, train-acc 0.865, loss 0.360, para-norm 5115.943, grad-norm 6.782, time 7.24s, 
epoch 179, batches 13000|18183, train-acc 0.857, loss 0.373, para-norm 5115.588, grad-norm 6.757, time 7.26s, 
epoch 179, batches 14000|18183, train-acc 0.861, loss 0.364, para-norm 5115.062, grad-norm 8.305, time 7.27s, 
epoch 179, batches 15000|18183, train-acc 0.863, loss 0.362, para-norm 5114.789, grad-norm 3.705, time 7.25s, 
epoch 179, batches 16000|18183, train-acc 0.859, loss 0.370, para-norm 5114.410, grad-norm 9.272, time 7.54s, 
epoch 179, batches 17000|18183, train-acc 0.858, loss 0.364, para-norm 5114.578, grad-norm 2.631, time 7.27s, 
epoch 179, batches 18000|18183, train-acc 0.861, loss 0.366, para-norm 5114.059, grad-norm 7.356, time 7.25s, 
epoch 179, batches 18183|18183, train-acc 0.865, loss 0.365, para-norm 5114.137, grad-norm 9.404, time 1.33s, 
dev-acc 0.860
epoch 180, batches 1000|18183, train-acc 0.865, loss 0.357, para-norm 5114.917, grad-norm 8.347, time 8.01s, 
epoch 180, batches 2000|18183, train-acc 0.863, loss 0.361, para-norm 5114.373, grad-norm 6.390, time 7.92s, 
epoch 180, batches 3000|18183, train-acc 0.863, loss 0.360, para-norm 5114.576, grad-norm 9.572, time 7.54s, 
epoch 180, batches 4000|18183, train-acc 0.862, loss 0.363, para-norm 5114.034, grad-norm 3.551, time 7.26s, 
epoch 180, batches 5000|18183, train-acc 0.861, loss 0.363, para-norm 5114.607, grad-norm 7.411, time 7.26s, 
epoch 180, batches 6000|18183, train-acc 0.863, loss 0.359, para-norm 5114.043, grad-norm 7.026, time 7.27s, 
epoch 180, batches 7000|18183, train-acc 0.861, loss 0.366, para-norm 5113.150, grad-norm 5.852, time 7.26s, 
epoch 180, batches 8000|18183, train-acc 0.859, loss 0.368, para-norm 5113.567, grad-norm 4.635, time 7.27s, 
epoch 180, batches 9000|18183, train-acc 0.863, loss 0.364, para-norm 5114.055, grad-norm 5.109, time 7.28s, 
epoch 180, batches 10000|18183, train-acc 0.861, loss 0.366, para-norm 5113.704, grad-norm 2.900, time 7.26s, 
epoch 180, batches 11000|18183, train-acc 0.864, loss 0.361, para-norm 5113.700, grad-norm 8.367, time 7.25s, 
epoch 180, batches 12000|18183, train-acc 0.858, loss 0.371, para-norm 5113.126, grad-norm 5.665, time 7.26s, 
epoch 180, batches 13000|18183, train-acc 0.862, loss 0.367, para-norm 5113.595, grad-norm 6.079, time 7.27s, 
epoch 180, batches 14000|18183, train-acc 0.862, loss 0.358, para-norm 5113.808, grad-norm 5.770, time 7.26s, 
epoch 180, batches 15000|18183, train-acc 0.861, loss 0.368, para-norm 5113.540, grad-norm 9.495, time 7.28s, 
epoch 180, batches 16000|18183, train-acc 0.862, loss 0.361, para-norm 5113.867, grad-norm 4.223, time 7.28s, 
epoch 180, batches 17000|18183, train-acc 0.864, loss 0.361, para-norm 5113.901, grad-norm 6.681, time 7.24s, 
epoch 180, batches 18000|18183, train-acc 0.864, loss 0.359, para-norm 5114.095, grad-norm 8.668, time 7.27s, 
epoch 180, batches 18183|18183, train-acc 0.854, loss 0.374, para-norm 5114.071, grad-norm 8.951, time 1.32s, 
dev-acc 0.858
epoch 181, batches 1000|18183, train-acc 0.862, loss 0.363, para-norm 5114.464, grad-norm 3.176, time 8.02s, 
epoch 181, batches 2000|18183, train-acc 0.859, loss 0.370, para-norm 5114.289, grad-norm 6.201, time 8.03s, 
epoch 181, batches 3000|18183, train-acc 0.863, loss 0.361, para-norm 5114.212, grad-norm 8.262, time 8.00s, 
epoch 181, batches 4000|18183, train-acc 0.863, loss 0.354, para-norm 5114.393, grad-norm 2.352, time 7.97s, 
epoch 181, batches 5000|18183, train-acc 0.863, loss 0.360, para-norm 5115.107, grad-norm 4.219, time 8.01s, 
epoch 181, batches 6000|18183, train-acc 0.860, loss 0.368, para-norm 5114.702, grad-norm 8.521, time 7.95s, 
epoch 181, batches 7000|18183, train-acc 0.864, loss 0.360, para-norm 5114.152, grad-norm 4.716, time 8.02s, 
epoch 181, batches 8000|18183, train-acc 0.861, loss 0.367, para-norm 5114.435, grad-norm 4.612, time 7.56s, 
epoch 181, batches 9000|18183, train-acc 0.864, loss 0.357, para-norm 5114.452, grad-norm 6.398, time 7.23s, 
epoch 181, batches 10000|18183, train-acc 0.865, loss 0.360, para-norm 5114.059, grad-norm 5.219, time 7.23s, 
epoch 181, batches 11000|18183, train-acc 0.861, loss 0.363, para-norm 5114.386, grad-norm 8.174, time 7.24s, 
epoch 181, batches 12000|18183, train-acc 0.859, loss 0.365, para-norm 5113.700, grad-norm 1.395, time 7.23s, 
epoch 181, batches 13000|18183, train-acc 0.862, loss 0.362, para-norm 5113.712, grad-norm 4.119, time 7.23s, 
epoch 181, batches 14000|18183, train-acc 0.861, loss 0.364, para-norm 5113.825, grad-norm 3.734, time 7.25s, 
epoch 181, batches 15000|18183, train-acc 0.860, loss 0.370, para-norm 5113.002, grad-norm 2.066, time 7.23s, 
epoch 181, batches 16000|18183, train-acc 0.861, loss 0.364, para-norm 5112.801, grad-norm 5.365, time 7.23s, 
epoch 181, batches 17000|18183, train-acc 0.861, loss 0.365, para-norm 5113.191, grad-norm 8.300, time 7.21s, 
epoch 181, batches 18000|18183, train-acc 0.857, loss 0.371, para-norm 5112.930, grad-norm 2.944, time 7.02s, 
epoch 181, batches 18183|18183, train-acc 0.852, loss 0.378, para-norm 5112.881, grad-norm 4.995, time 1.29s, 
dev-acc 0.861
epoch 182, batches 1000|18183, train-acc 0.865, loss 0.357, para-norm 5112.545, grad-norm 3.746, time 7.10s, 
epoch 182, batches 2000|18183, train-acc 0.864, loss 0.357, para-norm 5113.195, grad-norm 3.564, time 7.02s, 
epoch 182, batches 3000|18183, train-acc 0.859, loss 0.369, para-norm 5112.858, grad-norm 26.947, time 7.05s, 
epoch 182, batches 4000|18183, train-acc 0.864, loss 0.353, para-norm 5113.039, grad-norm 4.666, time 7.00s, 
epoch 182, batches 5000|18183, train-acc 0.867, loss 0.355, para-norm 5113.076, grad-norm 2.867, time 7.09s, 
epoch 182, batches 6000|18183, train-acc 0.859, loss 0.369, para-norm 5111.986, grad-norm 2.684, time 7.05s, 
epoch 182, batches 7000|18183, train-acc 0.866, loss 0.353, para-norm 5113.215, grad-norm 5.529, time 7.04s, 
epoch 182, batches 8000|18183, train-acc 0.865, loss 0.356, para-norm 5112.966, grad-norm 5.186, time 7.04s, 
epoch 182, batches 9000|18183, train-acc 0.858, loss 0.375, para-norm 5112.410, grad-norm 3.417, time 7.04s, 
epoch 182, batches 10000|18183, train-acc 0.863, loss 0.358, para-norm 5112.370, grad-norm 3.422, time 7.04s, 
epoch 182, batches 11000|18183, train-acc 0.858, loss 0.369, para-norm 5112.330, grad-norm 6.664, time 7.06s, 
epoch 182, batches 12000|18183, train-acc 0.860, loss 0.370, para-norm 5112.002, grad-norm 5.072, time 7.06s, 
epoch 182, batches 13000|18183, train-acc 0.862, loss 0.364, para-norm 5112.248, grad-norm 2.563, time 7.04s, 
epoch 182, batches 14000|18183, train-acc 0.861, loss 0.367, para-norm 5111.673, grad-norm 6.345, time 7.06s, 
epoch 182, batches 15000|18183, train-acc 0.863, loss 0.358, para-norm 5111.859, grad-norm 3.836, time 7.05s, 
epoch 182, batches 16000|18183, train-acc 0.861, loss 0.364, para-norm 5111.452, grad-norm 6.905, time 7.05s, 
epoch 182, batches 17000|18183, train-acc 0.862, loss 0.363, para-norm 5111.747, grad-norm 2.521, time 7.04s, 
epoch 182, batches 18000|18183, train-acc 0.861, loss 0.367, para-norm 5112.117, grad-norm 9.071, time 7.06s, 
epoch 182, batches 18183|18183, train-acc 0.864, loss 0.358, para-norm 5112.278, grad-norm 20.101, time 1.28s, 
dev-acc 0.859
epoch 183, batches 1000|18183, train-acc 0.862, loss 0.361, para-norm 5111.590, grad-norm 6.586, time 7.09s, 
epoch 183, batches 2000|18183, train-acc 0.860, loss 0.365, para-norm 5111.516, grad-norm 5.227, time 7.04s, 
epoch 183, batches 3000|18183, train-acc 0.861, loss 0.366, para-norm 5111.763, grad-norm 5.682, time 7.03s, 
epoch 183, batches 4000|18183, train-acc 0.861, loss 0.363, para-norm 5111.713, grad-norm 4.725, time 7.03s, 
epoch 183, batches 5000|18183, train-acc 0.865, loss 0.355, para-norm 5111.848, grad-norm 5.150, time 7.04s, 
epoch 183, batches 6000|18183, train-acc 0.859, loss 0.370, para-norm 5112.447, grad-norm 4.575, time 7.04s, 
epoch 183, batches 7000|18183, train-acc 0.860, loss 0.364, para-norm 5112.213, grad-norm 6.811, time 7.04s, 
epoch 183, batches 8000|18183, train-acc 0.863, loss 0.357, para-norm 5112.487, grad-norm 11.388, time 7.02s, 
epoch 183, batches 9000|18183, train-acc 0.863, loss 0.358, para-norm 5112.009, grad-norm 3.226, time 7.03s, 
epoch 183, batches 10000|18183, train-acc 0.860, loss 0.366, para-norm 5112.479, grad-norm 2.993, time 7.03s, 
epoch 183, batches 11000|18183, train-acc 0.866, loss 0.354, para-norm 5112.260, grad-norm 4.166, time 7.02s, 
epoch 183, batches 12000|18183, train-acc 0.860, loss 0.368, para-norm 5118.717, grad-norm 1.589, time 7.02s, 
epoch 183, batches 13000|18183, train-acc 0.863, loss 0.357, para-norm 5118.032, grad-norm 4.725, time 7.04s, 
epoch 183, batches 14000|18183, train-acc 0.861, loss 0.363, para-norm 5117.286, grad-norm 4.408, time 7.03s, 
epoch 183, batches 15000|18183, train-acc 0.861, loss 0.368, para-norm 5116.959, grad-norm 3.811, time 7.03s, 
epoch 183, batches 16000|18183, train-acc 0.861, loss 0.361, para-norm 5116.695, grad-norm 3.649, time 7.04s, 
epoch 183, batches 17000|18183, train-acc 0.860, loss 0.368, para-norm 5116.634, grad-norm 4.055, time 7.05s, 
epoch 183, batches 18000|18183, train-acc 0.865, loss 0.359, para-norm 5116.619, grad-norm 5.063, time 7.09s, 
epoch 183, batches 18183|18183, train-acc 0.854, loss 0.389, para-norm 5116.417, grad-norm 1.534, time 1.29s, 
dev-acc 0.862
epoch 184, batches 1000|18183, train-acc 0.859, loss 0.367, para-norm 5116.408, grad-norm 3.448, time 7.08s, 
epoch 184, batches 2000|18183, train-acc 0.862, loss 0.364, para-norm 5120.123, grad-norm 10.534, time 7.04s, 
epoch 184, batches 3000|18183, train-acc 0.863, loss 0.358, para-norm 5119.912, grad-norm 11.196, time 7.02s, 
epoch 184, batches 4000|18183, train-acc 0.867, loss 0.351, para-norm 5119.955, grad-norm 7.446, time 7.03s, 
epoch 184, batches 5000|18183, train-acc 0.860, loss 0.370, para-norm 5120.129, grad-norm 5.396, time 7.12s, 
epoch 184, batches 6000|18183, train-acc 0.861, loss 0.364, para-norm 5119.313, grad-norm 7.052, time 7.03s, 
epoch 184, batches 7000|18183, train-acc 0.866, loss 0.351, para-norm 5118.917, grad-norm 4.684, time 7.03s, 
epoch 184, batches 8000|18183, train-acc 0.861, loss 0.363, para-norm 5118.628, grad-norm 2.957, time 7.03s, 
epoch 184, batches 9000|18183, train-acc 0.861, loss 0.368, para-norm 5118.430, grad-norm 19.828, time 7.04s, 
epoch 184, batches 10000|18183, train-acc 0.861, loss 0.363, para-norm 5117.958, grad-norm 12.493, time 7.05s, 
epoch 184, batches 11000|18183, train-acc 0.862, loss 0.367, para-norm 5116.558, grad-norm 4.983, time 7.03s, 
epoch 184, batches 12000|18183, train-acc 0.864, loss 0.358, para-norm 5116.738, grad-norm 4.127, time 7.02s, 
epoch 184, batches 13000|18183, train-acc 0.858, loss 0.370, para-norm 5116.410, grad-norm 4.974, time 7.05s, 
epoch 184, batches 14000|18183, train-acc 0.863, loss 0.363, para-norm 5123.353, grad-norm 3.865, time 7.04s, 
epoch 184, batches 15000|18183, train-acc 0.860, loss 0.368, para-norm 5122.595, grad-norm 5.896, time 7.09s, 
epoch 184, batches 16000|18183, train-acc 0.862, loss 0.364, para-norm 5121.959, grad-norm 3.875, time 7.04s, 
epoch 184, batches 17000|18183, train-acc 0.865, loss 0.357, para-norm 5122.180, grad-norm 6.230, time 7.01s, 
epoch 184, batches 18000|18183, train-acc 0.861, loss 0.362, para-norm 5122.519, grad-norm 5.774, time 7.00s, 
epoch 184, batches 18183|18183, train-acc 0.869, loss 0.341, para-norm 5122.147, grad-norm 5.245, time 1.28s, 
dev-acc 0.861
epoch 185, batches 1000|18183, train-acc 0.860, loss 0.366, para-norm 5122.148, grad-norm 3.016, time 7.08s, 
epoch 185, batches 2000|18183, train-acc 0.866, loss 0.355, para-norm 5123.001, grad-norm 0.997, time 7.01s, 
epoch 185, batches 3000|18183, train-acc 0.864, loss 0.358, para-norm 5123.588, grad-norm 3.004, time 7.02s, 
epoch 185, batches 4000|18183, train-acc 0.863, loss 0.358, para-norm 5122.650, grad-norm 4.764, time 7.03s, 
epoch 185, batches 5000|18183, train-acc 0.869, loss 0.350, para-norm 5122.405, grad-norm 5.780, time 7.02s, 
epoch 185, batches 6000|18183, train-acc 0.862, loss 0.365, para-norm 5123.004, grad-norm 11.297, time 7.02s, 
epoch 185, batches 7000|18183, train-acc 0.859, loss 0.366, para-norm 5122.916, grad-norm 3.947, time 7.02s, 
epoch 185, batches 8000|18183, train-acc 0.864, loss 0.361, para-norm 5123.226, grad-norm 1.449, time 7.04s, 
epoch 185, batches 9000|18183, train-acc 0.857, loss 0.366, para-norm 5122.776, grad-norm 2.443, time 7.03s, 
epoch 185, batches 10000|18183, train-acc 0.862, loss 0.365, para-norm 5122.452, grad-norm 6.169, time 7.03s, 
epoch 185, batches 11000|18183, train-acc 0.864, loss 0.359, para-norm 5122.200, grad-norm 3.661, time 7.03s, 
epoch 185, batches 12000|18183, train-acc 0.860, loss 0.364, para-norm 5122.735, grad-norm 5.466, time 7.03s, 
epoch 185, batches 13000|18183, train-acc 0.863, loss 0.367, para-norm 5122.502, grad-norm 1.623, time 7.04s, 
epoch 185, batches 14000|18183, train-acc 0.867, loss 0.351, para-norm 5122.391, grad-norm 6.324, time 7.02s, 
epoch 185, batches 15000|18183, train-acc 0.861, loss 0.366, para-norm 5123.558, grad-norm 6.012, time 7.03s, 
epoch 185, batches 16000|18183, train-acc 0.862, loss 0.366, para-norm 5122.786, grad-norm 7.340, time 7.02s, 
epoch 185, batches 17000|18183, train-acc 0.864, loss 0.361, para-norm 5122.199, grad-norm 4.659, time 7.03s, 
epoch 185, batches 18000|18183, train-acc 0.861, loss 0.363, para-norm 5122.034, grad-norm 11.888, time 7.03s, 
epoch 185, batches 18183|18183, train-acc 0.856, loss 0.376, para-norm 5122.024, grad-norm 5.376, time 1.29s, 
dev-acc 0.863
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
	142 0.860
	147 0.862
	152 0.862
	185 0.863
save model!
epoch 186, batches 1000|18183, train-acc 0.863, loss 0.360, para-norm 5122.052, grad-norm 4.914, time 7.11s, 
epoch 186, batches 2000|18183, train-acc 0.862, loss 0.361, para-norm 5121.906, grad-norm 3.502, time 7.02s, 
epoch 186, batches 3000|18183, train-acc 0.865, loss 0.359, para-norm 5121.491, grad-norm 3.263, time 7.03s, 
epoch 186, batches 4000|18183, train-acc 0.865, loss 0.356, para-norm 5122.075, grad-norm 5.895, time 7.02s, 
epoch 186, batches 5000|18183, train-acc 0.864, loss 0.360, para-norm 5122.129, grad-norm 0.864, time 7.01s, 
epoch 186, batches 6000|18183, train-acc 0.865, loss 0.360, para-norm 5121.522, grad-norm 2.620, time 7.23s, 
epoch 186, batches 7000|18183, train-acc 0.864, loss 0.357, para-norm 5121.435, grad-norm 2.619, time 7.88s, 
epoch 186, batches 8000|18183, train-acc 0.861, loss 0.362, para-norm 5121.459, grad-norm 7.825, time 7.89s, 
epoch 186, batches 9000|18183, train-acc 0.859, loss 0.370, para-norm 5120.228, grad-norm 22.371, time 7.89s, 
epoch 186, batches 10000|18183, train-acc 0.864, loss 0.359, para-norm 5120.514, grad-norm 4.114, time 7.89s, 
epoch 186, batches 11000|18183, train-acc 0.860, loss 0.367, para-norm 5119.566, grad-norm 0.218, time 7.88s, 
epoch 186, batches 12000|18183, train-acc 0.863, loss 0.359, para-norm 5119.630, grad-norm 3.875, time 7.88s, 
epoch 186, batches 13000|18183, train-acc 0.862, loss 0.364, para-norm 5119.174, grad-norm 3.895, time 7.88s, 
epoch 186, batches 14000|18183, train-acc 0.862, loss 0.361, para-norm 5119.470, grad-norm 3.752, time 7.88s, 
epoch 186, batches 15000|18183, train-acc 0.861, loss 0.367, para-norm 5119.175, grad-norm 8.082, time 7.88s, 
epoch 186, batches 16000|18183, train-acc 0.863, loss 0.357, para-norm 5119.904, grad-norm 4.433, time 7.88s, 
epoch 186, batches 17000|18183, train-acc 0.862, loss 0.365, para-norm 5120.290, grad-norm 2.479, time 7.88s, 
epoch 186, batches 18000|18183, train-acc 0.866, loss 0.354, para-norm 5119.918, grad-norm 5.469, time 7.87s, 
epoch 186, batches 18183|18183, train-acc 0.861, loss 0.377, para-norm 5120.061, grad-norm 4.700, time 1.45s, 
dev-acc 0.860
epoch 187, batches 1000|18183, train-acc 0.865, loss 0.358, para-norm 5120.627, grad-norm 6.444, time 8.02s, 
epoch 187, batches 2000|18183, train-acc 0.861, loss 0.361, para-norm 5125.822, grad-norm 6.996, time 7.87s, 
epoch 187, batches 3000|18183, train-acc 0.863, loss 0.358, para-norm 5127.005, grad-norm 2.694, time 7.87s, 
epoch 187, batches 4000|18183, train-acc 0.862, loss 0.361, para-norm 5126.528, grad-norm 4.024, time 7.88s, 
epoch 187, batches 5000|18183, train-acc 0.866, loss 0.353, para-norm 5126.622, grad-norm 3.359, time 7.88s, 
epoch 187, batches 6000|18183, train-acc 0.862, loss 0.364, para-norm 5125.965, grad-norm 7.858, time 7.87s, 
epoch 187, batches 7000|18183, train-acc 0.863, loss 0.360, para-norm 5125.925, grad-norm 0.214, time 7.88s, 
epoch 187, batches 8000|18183, train-acc 0.862, loss 0.373, para-norm 5124.559, grad-norm 5.387, time 7.90s, 
epoch 187, batches 9000|18183, train-acc 0.864, loss 0.356, para-norm 5124.769, grad-norm 162.873, time 7.90s, 
epoch 187, batches 10000|18183, train-acc 0.861, loss 0.362, para-norm 5124.117, grad-norm 4.686, time 7.89s, 
epoch 187, batches 11000|18183, train-acc 0.864, loss 0.363, para-norm 5124.308, grad-norm 2.232, time 7.88s, 
epoch 187, batches 12000|18183, train-acc 0.863, loss 0.359, para-norm 5123.812, grad-norm 5.327, time 7.88s, 
epoch 187, batches 13000|18183, train-acc 0.867, loss 0.355, para-norm 5122.872, grad-norm 5.042, time 7.88s, 
epoch 187, batches 14000|18183, train-acc 0.861, loss 0.366, para-norm 5122.160, grad-norm 5.350, time 7.87s, 
epoch 187, batches 15000|18183, train-acc 0.864, loss 0.357, para-norm 5122.196, grad-norm 9.007, time 7.87s, 
epoch 187, batches 16000|18183, train-acc 0.863, loss 0.359, para-norm 5121.157, grad-norm 6.552, time 7.88s, 
epoch 187, batches 17000|18183, train-acc 0.861, loss 0.365, para-norm 5121.499, grad-norm 6.974, time 7.88s, 
epoch 187, batches 18000|18183, train-acc 0.863, loss 0.359, para-norm 5121.421, grad-norm 8.653, time 7.88s, 
epoch 187, batches 18183|18183, train-acc 0.861, loss 0.368, para-norm 5121.049, grad-norm 5.152, time 1.44s, 
dev-acc 0.862
epoch 188, batches 1000|18183, train-acc 0.866, loss 0.353, para-norm 5121.363, grad-norm 5.793, time 7.94s, 
epoch 188, batches 2000|18183, train-acc 0.866, loss 0.350, para-norm 5122.133, grad-norm 2.753, time 7.87s, 
epoch 188, batches 3000|18183, train-acc 0.863, loss 0.363, para-norm 5121.661, grad-norm 94.419, time 7.89s, 
epoch 188, batches 4000|18183, train-acc 0.861, loss 0.363, para-norm 5121.232, grad-norm 6.504, time 7.88s, 
epoch 188, batches 5000|18183, train-acc 0.864, loss 0.357, para-norm 5121.033, grad-norm 6.390, time 7.88s, 
epoch 188, batches 6000|18183, train-acc 0.863, loss 0.360, para-norm 5121.463, grad-norm 10.473, time 7.30s, 
epoch 188, batches 7000|18183, train-acc 0.863, loss 0.361, para-norm 5120.886, grad-norm 0.000, time 7.04s, 
epoch 188, batches 8000|18183, train-acc 0.861, loss 0.366, para-norm 5120.520, grad-norm 7.352, time 7.03s, 
epoch 188, batches 9000|18183, train-acc 0.865, loss 0.355, para-norm 5119.985, grad-norm 5.245, time 7.03s, 
epoch 188, batches 10000|18183, train-acc 0.864, loss 0.362, para-norm 5119.780, grad-norm 5.528, time 7.04s, 
epoch 188, batches 11000|18183, train-acc 0.863, loss 0.361, para-norm 5119.701, grad-norm 10.131, time 7.05s, 
epoch 188, batches 12000|18183, train-acc 0.860, loss 0.367, para-norm 5119.223, grad-norm 5.882, time 7.05s, 
epoch 188, batches 13000|18183, train-acc 0.863, loss 0.360, para-norm 5118.641, grad-norm 5.614, time 7.06s, 
epoch 188, batches 14000|18183, train-acc 0.860, loss 0.368, para-norm 5118.088, grad-norm 3.850, time 7.05s, 
epoch 188, batches 15000|18183, train-acc 0.863, loss 0.358, para-norm 5117.604, grad-norm 3.122, time 7.04s, 
epoch 188, batches 16000|18183, train-acc 0.863, loss 0.359, para-norm 5117.418, grad-norm 3.356, time 7.04s, 
epoch 188, batches 17000|18183, train-acc 0.857, loss 0.367, para-norm 5117.216, grad-norm 5.391, time 7.05s, 
epoch 188, batches 18000|18183, train-acc 0.862, loss 0.364, para-norm 5117.222, grad-norm 3.926, time 7.05s, 
epoch 188, batches 18183|18183, train-acc 0.858, loss 0.370, para-norm 5117.071, grad-norm 9.863, time 1.29s, 
dev-acc 0.861
epoch 189, batches 1000|18183, train-acc 0.864, loss 0.355, para-norm 5117.662, grad-norm 6.133, time 7.20s, 
epoch 189, batches 2000|18183, train-acc 0.862, loss 0.361, para-norm 5118.183, grad-norm 2.798, time 7.05s, 
epoch 189, batches 3000|18183, train-acc 0.867, loss 0.347, para-norm 5118.480, grad-norm 4.109, time 7.03s, 
epoch 189, batches 4000|18183, train-acc 0.865, loss 0.354, para-norm 5117.967, grad-norm 5.765, time 7.17s, 
epoch 189, batches 5000|18183, train-acc 0.864, loss 0.357, para-norm 5118.137, grad-norm 182.868, time 7.05s, 
epoch 189, batches 6000|18183, train-acc 0.859, loss 0.367, para-norm 5118.185, grad-norm 5.221, time 7.05s, 
epoch 189, batches 7000|18183, train-acc 0.863, loss 0.363, para-norm 5117.802, grad-norm 8.822, time 7.04s, 
epoch 189, batches 8000|18183, train-acc 0.863, loss 0.360, para-norm 5117.310, grad-norm 1.934, time 7.04s, 
epoch 189, batches 9000|18183, train-acc 0.860, loss 0.363, para-norm 5116.892, grad-norm 4.587, time 7.05s, 
epoch 189, batches 10000|18183, train-acc 0.862, loss 0.359, para-norm 5117.027, grad-norm 0.005, time 7.04s, 
epoch 189, batches 11000|18183, train-acc 0.865, loss 0.357, para-norm 5117.468, grad-norm 3.665, time 7.02s, 
epoch 189, batches 12000|18183, train-acc 0.857, loss 0.372, para-norm 5116.687, grad-norm 19.695, time 7.07s, 
epoch 189, batches 13000|18183, train-acc 0.859, loss 0.366, para-norm 5117.232, grad-norm 5.838, time 7.08s, 
epoch 189, batches 14000|18183, train-acc 0.861, loss 0.360, para-norm 5120.983, grad-norm 7.439, time 7.20s, 
epoch 189, batches 15000|18183, train-acc 0.864, loss 0.358, para-norm 5120.730, grad-norm 6.600, time 7.88s, 
epoch 189, batches 16000|18183, train-acc 0.862, loss 0.361, para-norm 5120.338, grad-norm 5.228, time 7.88s, 
epoch 189, batches 17000|18183, train-acc 0.865, loss 0.358, para-norm 5120.105, grad-norm 1.888, time 7.87s, 
epoch 189, batches 18000|18183, train-acc 0.861, loss 0.368, para-norm 5120.084, grad-norm 3.520, time 7.87s, 
epoch 189, batches 18183|18183, train-acc 0.861, loss 0.368, para-norm 5119.982, grad-norm 1.717, time 1.44s, 
dev-acc 0.860
epoch 190, batches 1000|18183, train-acc 0.865, loss 0.355, para-norm 5120.215, grad-norm 2.754, time 7.95s, 
epoch 190, batches 2000|18183, train-acc 0.865, loss 0.354, para-norm 5120.256, grad-norm 2.588, time 7.87s, 
epoch 190, batches 3000|18183, train-acc 0.864, loss 0.353, para-norm 5119.380, grad-norm 4.995, time 7.87s, 
epoch 190, batches 4000|18183, train-acc 0.867, loss 0.353, para-norm 5119.352, grad-norm 5.631, time 7.87s, 
epoch 190, batches 5000|18183, train-acc 0.863, loss 0.356, para-norm 5119.280, grad-norm 8.297, time 7.87s, 
epoch 190, batches 6000|18183, train-acc 0.865, loss 0.361, para-norm 5119.201, grad-norm 5.135, time 7.89s, 
epoch 190, batches 7000|18183, train-acc 0.863, loss 0.357, para-norm 5119.202, grad-norm 4.700, time 7.88s, 
epoch 190, batches 8000|18183, train-acc 0.864, loss 0.360, para-norm 5119.082, grad-norm 9.648, time 7.88s, 
epoch 190, batches 9000|18183, train-acc 0.860, loss 0.368, para-norm 5118.653, grad-norm 14.786, time 7.87s, 
epoch 190, batches 10000|18183, train-acc 0.861, loss 0.362, para-norm 5118.650, grad-norm 6.194, time 7.88s, 
epoch 190, batches 11000|18183, train-acc 0.863, loss 0.362, para-norm 5118.669, grad-norm 5.859, time 7.87s, 
epoch 190, batches 12000|18183, train-acc 0.860, loss 0.372, para-norm 5118.965, grad-norm 6.320, time 7.89s, 
epoch 190, batches 13000|18183, train-acc 0.864, loss 0.358, para-norm 5118.699, grad-norm 6.426, time 7.88s, 
epoch 190, batches 14000|18183, train-acc 0.866, loss 0.354, para-norm 5118.754, grad-norm 3.587, time 7.87s, 
epoch 190, batches 15000|18183, train-acc 0.862, loss 0.359, para-norm 5119.079, grad-norm 6.402, time 7.87s, 
epoch 190, batches 16000|18183, train-acc 0.863, loss 0.360, para-norm 5119.429, grad-norm 2.789, time 7.88s, 
epoch 190, batches 17000|18183, train-acc 0.863, loss 0.361, para-norm 5118.745, grad-norm 6.354, time 7.88s, 
epoch 190, batches 18000|18183, train-acc 0.862, loss 0.361, para-norm 5119.041, grad-norm 2.513, time 7.99s, 
epoch 190, batches 18183|18183, train-acc 0.853, loss 0.384, para-norm 5119.045, grad-norm 4.753, time 1.45s, 
dev-acc 0.862
epoch 191, batches 1000|18183, train-acc 0.862, loss 0.364, para-norm 5118.508, grad-norm 8.539, time 7.91s, 
epoch 191, batches 2000|18183, train-acc 0.868, loss 0.346, para-norm 5118.899, grad-norm 5.700, time 7.86s, 
epoch 191, batches 3000|18183, train-acc 0.861, loss 0.363, para-norm 5118.360, grad-norm 4.592, time 7.88s, 
epoch 191, batches 4000|18183, train-acc 0.865, loss 0.361, para-norm 5118.377, grad-norm 5.893, time 7.90s, 
epoch 191, batches 5000|18183, train-acc 0.864, loss 0.361, para-norm 5117.553, grad-norm 6.947, time 7.87s, 
epoch 191, batches 6000|18183, train-acc 0.867, loss 0.352, para-norm 5117.837, grad-norm 3.774, time 7.86s, 
epoch 191, batches 7000|18183, train-acc 0.860, loss 0.362, para-norm 5118.217, grad-norm 6.454, time 7.89s, 
epoch 191, batches 8000|18183, train-acc 0.867, loss 0.355, para-norm 5117.649, grad-norm 6.159, time 7.88s, 
epoch 191, batches 9000|18183, train-acc 0.862, loss 0.364, para-norm 5116.897, grad-norm 4.994, time 7.70s, 
epoch 191, batches 10000|18183, train-acc 0.861, loss 0.365, para-norm 5116.809, grad-norm 6.896, time 7.65s, 
epoch 191, batches 11000|18183, train-acc 0.862, loss 0.358, para-norm 5117.129, grad-norm 3.142, time 7.65s, 
epoch 191, batches 12000|18183, train-acc 0.864, loss 0.359, para-norm 5116.511, grad-norm 6.213, time 7.40s, 
epoch 191, batches 13000|18183, train-acc 0.861, loss 0.371, para-norm 5116.721, grad-norm 3.676, time 7.38s, 
epoch 191, batches 14000|18183, train-acc 0.861, loss 0.362, para-norm 5116.452, grad-norm 6.316, time 7.40s, 
epoch 191, batches 15000|18183, train-acc 0.864, loss 0.358, para-norm 5116.989, grad-norm 3.673, time 7.28s, 
epoch 191, batches 16000|18183, train-acc 0.863, loss 0.358, para-norm 5116.288, grad-norm 5.432, time 7.27s, 
epoch 191, batches 17000|18183, train-acc 0.863, loss 0.362, para-norm 5115.551, grad-norm 6.223, time 7.27s, 
epoch 191, batches 18000|18183, train-acc 0.863, loss 0.360, para-norm 5115.348, grad-norm 3.096, time 7.25s, 
epoch 191, batches 18183|18183, train-acc 0.859, loss 0.385, para-norm 5115.400, grad-norm 7.175, time 1.33s, 
dev-acc 0.860
epoch 192, batches 1000|18183, train-acc 0.866, loss 0.352, para-norm 5115.666, grad-norm 19.789, time 7.32s, 
epoch 192, batches 2000|18183, train-acc 0.862, loss 0.362, para-norm 5115.818, grad-norm 1.278, time 7.27s, 
epoch 192, batches 3000|18183, train-acc 0.862, loss 0.360, para-norm 5115.209, grad-norm 9.112, time 7.27s, 
epoch 192, batches 4000|18183, train-acc 0.864, loss 0.358, para-norm 5115.345, grad-norm 11.387, time 7.27s, 
epoch 192, batches 5000|18183, train-acc 0.866, loss 0.357, para-norm 5115.667, grad-norm 6.933, time 7.26s, 
epoch 192, batches 6000|18183, train-acc 0.860, loss 0.368, para-norm 5114.960, grad-norm 11.144, time 7.28s, 
epoch 192, batches 7000|18183, train-acc 0.864, loss 0.362, para-norm 5115.199, grad-norm 6.911, time 7.26s, 
epoch 192, batches 8000|18183, train-acc 0.862, loss 0.364, para-norm 5115.195, grad-norm 2.626, time 7.28s, 
epoch 192, batches 9000|18183, train-acc 0.861, loss 0.369, para-norm 5114.787, grad-norm 2.146, time 7.26s, 
epoch 192, batches 10000|18183, train-acc 0.864, loss 0.356, para-norm 5114.715, grad-norm 6.794, time 7.26s, 
epoch 192, batches 11000|18183, train-acc 0.862, loss 0.365, para-norm 5114.607, grad-norm 7.520, time 7.26s, 
epoch 192, batches 12000|18183, train-acc 0.864, loss 0.352, para-norm 5114.706, grad-norm 3.421, time 7.27s, 
epoch 192, batches 13000|18183, train-acc 0.866, loss 0.353, para-norm 5114.527, grad-norm 9.400, time 7.61s, 
epoch 192, batches 14000|18183, train-acc 0.860, loss 0.366, para-norm 5113.928, grad-norm 4.722, time 7.61s, 
epoch 192, batches 15000|18183, train-acc 0.862, loss 0.363, para-norm 5114.171, grad-norm 1.095, time 7.60s, 
epoch 192, batches 16000|18183, train-acc 0.862, loss 0.359, para-norm 5114.591, grad-norm 6.215, time 7.60s, 
epoch 192, batches 17000|18183, train-acc 0.863, loss 0.360, para-norm 5114.665, grad-norm 4.730, time 7.78s, 
epoch 192, batches 18000|18183, train-acc 0.866, loss 0.355, para-norm 5114.869, grad-norm 18.092, time 7.59s, 
epoch 192, batches 18183|18183, train-acc 0.853, loss 0.375, para-norm 5115.061, grad-norm 6.417, time 1.40s, 
dev-acc 0.862
epoch 193, batches 1000|18183, train-acc 0.868, loss 0.349, para-norm 5115.017, grad-norm 5.008, time 7.64s, 
epoch 193, batches 2000|18183, train-acc 0.864, loss 0.359, para-norm 5115.453, grad-norm 2.554, time 7.83s, 
epoch 193, batches 3000|18183, train-acc 0.864, loss 0.360, para-norm 5115.313, grad-norm 3.496, time 7.60s, 
epoch 193, batches 4000|18183, train-acc 0.861, loss 0.366, para-norm 5115.405, grad-norm 6.075, time 7.61s, 
epoch 193, batches 5000|18183, train-acc 0.863, loss 0.357, para-norm 5115.898, grad-norm 1.692, time 7.60s, 
epoch 193, batches 6000|18183, train-acc 0.866, loss 0.358, para-norm 5115.801, grad-norm 6.451, time 7.58s, 
epoch 193, batches 7000|18183, train-acc 0.864, loss 0.356, para-norm 5115.629, grad-norm 4.615, time 7.60s, 
epoch 193, batches 8000|18183, train-acc 0.864, loss 0.360, para-norm 5115.561, grad-norm 15.615, time 7.60s, 
epoch 193, batches 9000|18183, train-acc 0.862, loss 0.361, para-norm 5115.292, grad-norm 3.051, time 7.62s, 
epoch 193, batches 10000|18183, train-acc 0.861, loss 0.363, para-norm 5115.284, grad-norm 4.564, time 7.61s, 
epoch 193, batches 11000|18183, train-acc 0.863, loss 0.360, para-norm 5114.889, grad-norm 14.458, time 7.62s, 
epoch 193, batches 12000|18183, train-acc 0.864, loss 0.356, para-norm 5115.646, grad-norm 0.002, time 7.60s, 
epoch 193, batches 13000|18183, train-acc 0.864, loss 0.359, para-norm 5115.248, grad-norm 6.424, time 7.60s, 
epoch 193, batches 14000|18183, train-acc 0.860, loss 0.360, para-norm 5115.784, grad-norm 3.290, time 7.60s, 
epoch 193, batches 15000|18183, train-acc 0.860, loss 0.367, para-norm 5121.873, grad-norm 0.016, time 7.61s, 
epoch 193, batches 16000|18183, train-acc 0.864, loss 0.356, para-norm 5121.806, grad-norm 0.077, time 7.61s, 
epoch 193, batches 17000|18183, train-acc 0.865, loss 0.357, para-norm 5121.721, grad-norm 4.447, time 7.61s, 
epoch 193, batches 18000|18183, train-acc 0.866, loss 0.351, para-norm 5121.357, grad-norm 3.745, time 7.33s, 
epoch 193, batches 18183|18183, train-acc 0.864, loss 0.360, para-norm 5121.368, grad-norm 5.890, time 1.33s, 
dev-acc 0.858
epoch 194, batches 1000|18183, train-acc 0.868, loss 0.352, para-norm 5121.872, grad-norm 5.092, time 7.38s, 
epoch 194, batches 2000|18183, train-acc 0.863, loss 0.363, para-norm 5121.781, grad-norm 5.841, time 7.27s, 
epoch 194, batches 3000|18183, train-acc 0.866, loss 0.354, para-norm 5121.183, grad-norm 2.421, time 7.35s, 
epoch 194, batches 4000|18183, train-acc 0.861, loss 0.366, para-norm 5121.168, grad-norm 4.950, time 7.27s, 
epoch 194, batches 5000|18183, train-acc 0.867, loss 0.351, para-norm 5121.488, grad-norm 3.966, time 7.24s, 
epoch 194, batches 6000|18183, train-acc 0.861, loss 0.364, para-norm 5120.897, grad-norm 5.866, time 7.26s, 
epoch 194, batches 7000|18183, train-acc 0.864, loss 0.356, para-norm 5121.076, grad-norm 3.341, time 7.25s, 
epoch 194, batches 8000|18183, train-acc 0.863, loss 0.361, para-norm 5120.933, grad-norm 6.836, time 7.26s, 
epoch 194, batches 9000|18183, train-acc 0.863, loss 0.363, para-norm 5120.857, grad-norm 3.563, time 7.25s, 
epoch 194, batches 10000|18183, train-acc 0.865, loss 0.359, para-norm 5121.839, grad-norm 2.275, time 7.25s, 
epoch 194, batches 11000|18183, train-acc 0.865, loss 0.358, para-norm 5122.163, grad-norm 17.994, time 7.26s, 
epoch 194, batches 12000|18183, train-acc 0.862, loss 0.365, para-norm 5121.701, grad-norm 2.990, time 7.26s, 
epoch 194, batches 13000|18183, train-acc 0.864, loss 0.355, para-norm 5121.093, grad-norm 2.531, time 7.26s, 
epoch 194, batches 14000|18183, train-acc 0.862, loss 0.364, para-norm 5120.885, grad-norm 5.865, time 7.24s, 
epoch 194, batches 15000|18183, train-acc 0.865, loss 0.360, para-norm 5121.447, grad-norm 1.760, time 7.27s, 
epoch 194, batches 16000|18183, train-acc 0.862, loss 0.359, para-norm 5120.799, grad-norm 5.190, time 7.26s, 
epoch 194, batches 17000|18183, train-acc 0.862, loss 0.367, para-norm 5120.837, grad-norm 2.486, time 7.26s, 
epoch 194, batches 18000|18183, train-acc 0.862, loss 0.358, para-norm 5120.660, grad-norm 5.701, time 7.25s, 
epoch 194, batches 18183|18183, train-acc 0.866, loss 0.357, para-norm 5120.655, grad-norm 5.086, time 1.33s, 
dev-acc 0.861
epoch 195, batches 1000|18183, train-acc 0.868, loss 0.352, para-norm 5120.329, grad-norm 4.655, time 7.32s, 
epoch 195, batches 2000|18183, train-acc 0.861, loss 0.366, para-norm 5120.171, grad-norm 1093.021, time 7.27s, 
epoch 195, batches 3000|18183, train-acc 0.866, loss 0.354, para-norm 5120.334, grad-norm 4.273, time 7.26s, 
epoch 195, batches 4000|18183, train-acc 0.861, loss 0.364, para-norm 5119.764, grad-norm 8.431, time 7.27s, 
epoch 195, batches 5000|18183, train-acc 0.863, loss 0.360, para-norm 5119.460, grad-norm 2.147, time 7.25s, 
epoch 195, batches 6000|18183, train-acc 0.864, loss 0.356, para-norm 5119.592, grad-norm 5.191, time 7.26s, 
epoch 195, batches 7000|18183, train-acc 0.864, loss 0.356, para-norm 5119.796, grad-norm 5.801, time 7.25s, 
epoch 195, batches 8000|18183, train-acc 0.864, loss 0.360, para-norm 5119.651, grad-norm 8.850, time 7.26s, 
epoch 195, batches 9000|18183, train-acc 0.861, loss 0.363, para-norm 5119.641, grad-norm 5.218, time 7.27s, 
epoch 195, batches 10000|18183, train-acc 0.865, loss 0.356, para-norm 5119.493, grad-norm 5.337, time 7.25s, 
epoch 195, batches 11000|18183, train-acc 0.865, loss 0.356, para-norm 5119.721, grad-norm 0.861, time 7.28s, 
epoch 195, batches 12000|18183, train-acc 0.863, loss 0.363, para-norm 5118.618, grad-norm 3.795, time 7.26s, 
epoch 195, batches 13000|18183, train-acc 0.862, loss 0.361, para-norm 5118.882, grad-norm 8.780, time 7.25s, 
epoch 195, batches 14000|18183, train-acc 0.869, loss 0.351, para-norm 5118.773, grad-norm 5.388, time 7.25s, 
epoch 195, batches 15000|18183, train-acc 0.864, loss 0.360, para-norm 5118.098, grad-norm 5.465, time 7.26s, 
epoch 195, batches 16000|18183, train-acc 0.865, loss 0.354, para-norm 5118.420, grad-norm 2.283, time 7.26s, 
epoch 195, batches 17000|18183, train-acc 0.863, loss 0.359, para-norm 5118.323, grad-norm 7.019, time 7.26s, 
epoch 195, batches 18000|18183, train-acc 0.859, loss 0.372, para-norm 5117.840, grad-norm 3.796, time 7.27s, 
epoch 195, batches 18183|18183, train-acc 0.857, loss 0.370, para-norm 5117.594, grad-norm 4.380, time 1.33s, 
dev-acc 0.860
epoch 196, batches 1000|18183, train-acc 0.864, loss 0.356, para-norm 5117.526, grad-norm 5.102, time 7.44s, 
epoch 196, batches 2000|18183, train-acc 0.868, loss 0.349, para-norm 5124.148, grad-norm 0.024, time 7.25s, 
epoch 196, batches 3000|18183, train-acc 0.865, loss 0.354, para-norm 5124.237, grad-norm 7.563, time 7.26s, 
epoch 196, batches 4000|18183, train-acc 0.866, loss 0.358, para-norm 5124.479, grad-norm 9.931, time 7.25s, 
epoch 196, batches 5000|18183, train-acc 0.864, loss 0.360, para-norm 5124.313, grad-norm 6.305, time 7.37s, 
epoch 196, batches 6000|18183, train-acc 0.860, loss 0.369, para-norm 5128.828, grad-norm 6.402, time 7.28s, 
epoch 196, batches 7000|18183, train-acc 0.862, loss 0.362, para-norm 5130.133, grad-norm 3.332, time 7.26s, 
epoch 196, batches 8000|18183, train-acc 0.862, loss 0.359, para-norm 5129.656, grad-norm 1.826, time 7.27s, 
epoch 196, batches 9000|18183, train-acc 0.861, loss 0.362, para-norm 5129.224, grad-norm 4.285, time 7.27s, 
epoch 196, batches 10000|18183, train-acc 0.864, loss 0.360, para-norm 5128.467, grad-norm 6.805, time 7.27s, 
epoch 196, batches 11000|18183, train-acc 0.865, loss 0.355, para-norm 5128.114, grad-norm 4.859, time 7.27s, 
epoch 196, batches 12000|18183, train-acc 0.863, loss 0.361, para-norm 5127.967, grad-norm 3.312, time 7.28s, 
epoch 196, batches 13000|18183, train-acc 0.863, loss 0.358, para-norm 5128.132, grad-norm 2.738, time 7.26s, 
epoch 196, batches 14000|18183, train-acc 0.867, loss 0.353, para-norm 5127.321, grad-norm 2.684, time 7.25s, 
epoch 196, batches 15000|18183, train-acc 0.862, loss 0.360, para-norm 5126.674, grad-norm 3.400, time 7.26s, 
epoch 196, batches 16000|18183, train-acc 0.863, loss 0.358, para-norm 5126.155, grad-norm 10.126, time 7.25s, 
epoch 196, batches 17000|18183, train-acc 0.866, loss 0.355, para-norm 5126.342, grad-norm 13.220, time 7.25s, 
epoch 196, batches 18000|18183, train-acc 0.864, loss 0.357, para-norm 5126.389, grad-norm 3.858, time 7.27s, 
epoch 196, batches 18183|18183, train-acc 0.867, loss 0.351, para-norm 5126.302, grad-norm 3.272, time 1.33s, 
dev-acc 0.860
epoch 197, batches 1000|18183, train-acc 0.859, loss 0.365, para-norm 5125.940, grad-norm 5.264, time 7.35s, 
epoch 197, batches 2000|18183, train-acc 0.867, loss 0.353, para-norm 5126.209, grad-norm 3.770, time 7.41s, 
epoch 197, batches 3000|18183, train-acc 0.867, loss 0.352, para-norm 5125.478, grad-norm 5.706, time 7.26s, 
epoch 197, batches 4000|18183, train-acc 0.862, loss 0.356, para-norm 5124.657, grad-norm 6.227, time 7.26s, 
epoch 197, batches 5000|18183, train-acc 0.862, loss 0.356, para-norm 5124.654, grad-norm 4.615, time 7.27s, 
epoch 197, batches 6000|18183, train-acc 0.856, loss 0.371, para-norm 5124.149, grad-norm 5.499, time 7.37s, 
epoch 197, batches 7000|18183, train-acc 0.863, loss 0.359, para-norm 5124.275, grad-norm 7.225, time 7.28s, 
epoch 197, batches 8000|18183, train-acc 0.865, loss 0.353, para-norm 5124.425, grad-norm 12.779, time 7.26s, 
epoch 197, batches 9000|18183, train-acc 0.865, loss 0.354, para-norm 5124.690, grad-norm 4.115, time 7.27s, 
epoch 197, batches 10000|18183, train-acc 0.864, loss 0.356, para-norm 5124.722, grad-norm 3.826, time 7.28s, 
epoch 197, batches 11000|18183, train-acc 0.865, loss 0.359, para-norm 5125.034, grad-norm 8.572, time 7.25s, 
epoch 197, batches 12000|18183, train-acc 0.861, loss 0.364, para-norm 5124.999, grad-norm 6.728, time 7.28s, 
epoch 197, batches 13000|18183, train-acc 0.862, loss 0.361, para-norm 5124.887, grad-norm 7.238, time 7.27s, 
epoch 197, batches 14000|18183, train-acc 0.867, loss 0.357, para-norm 5124.391, grad-norm 1.996, time 7.27s, 
epoch 197, batches 15000|18183, train-acc 0.860, loss 0.366, para-norm 5123.137, grad-norm 1.926, time 7.27s, 
epoch 197, batches 16000|18183, train-acc 0.865, loss 0.357, para-norm 5122.481, grad-norm 5.688, time 7.26s, 
epoch 197, batches 17000|18183, train-acc 0.863, loss 0.360, para-norm 5122.812, grad-norm 10.104, time 7.26s, 
epoch 197, batches 18000|18183, train-acc 0.868, loss 0.356, para-norm 5122.116, grad-norm 3.688, time 7.27s, 
epoch 197, batches 18183|18183, train-acc 0.861, loss 0.372, para-norm 5121.962, grad-norm 5.874, time 1.33s, 
dev-acc 0.860
epoch 198, batches 1000|18183, train-acc 0.867, loss 0.351, para-norm 5122.505, grad-norm 4.973, time 7.31s, 
epoch 198, batches 2000|18183, train-acc 0.864, loss 0.362, para-norm 5122.606, grad-norm 3.181, time 7.28s, 
epoch 198, batches 3000|18183, train-acc 0.866, loss 0.351, para-norm 5122.606, grad-norm 2.480, time 7.28s, 
epoch 198, batches 4000|18183, train-acc 0.865, loss 0.354, para-norm 5121.868, grad-norm 4.894, time 7.27s, 
epoch 198, batches 5000|18183, train-acc 0.864, loss 0.359, para-norm 5122.173, grad-norm 8.389, time 7.27s, 
epoch 198, batches 6000|18183, train-acc 0.858, loss 0.366, para-norm 5121.918, grad-norm 17.294, time 7.27s, 
epoch 198, batches 7000|18183, train-acc 0.858, loss 0.367, para-norm 5122.531, grad-norm 10.593, time 7.27s, 
epoch 198, batches 8000|18183, train-acc 0.863, loss 0.356, para-norm 5121.877, grad-norm 4.287, time 7.28s, 
epoch 198, batches 9000|18183, train-acc 0.865, loss 0.356, para-norm 5121.132, grad-norm 7.790, time 7.27s, 
epoch 198, batches 10000|18183, train-acc 0.865, loss 0.354, para-norm 5121.004, grad-norm 5.241, time 7.27s, 
epoch 198, batches 11000|18183, train-acc 0.867, loss 0.349, para-norm 5121.312, grad-norm 7.080, time 7.27s, 
epoch 198, batches 12000|18183, train-acc 0.862, loss 0.361, para-norm 5120.431, grad-norm 1.563, time 7.28s, 
epoch 198, batches 13000|18183, train-acc 0.863, loss 0.363, para-norm 5120.146, grad-norm 5.532, time 7.27s, 
epoch 198, batches 14000|18183, train-acc 0.859, loss 0.367, para-norm 5120.417, grad-norm 5.386, time 7.28s, 
epoch 198, batches 15000|18183, train-acc 0.863, loss 0.360, para-norm 5119.782, grad-norm 10.744, time 7.29s, 
epoch 198, batches 16000|18183, train-acc 0.866, loss 0.354, para-norm 5119.780, grad-norm 4.088, time 7.37s, 
epoch 198, batches 17000|18183, train-acc 0.860, loss 0.368, para-norm 5119.339, grad-norm 4.116, time 7.26s, 
epoch 198, batches 18000|18183, train-acc 0.870, loss 0.350, para-norm 5119.522, grad-norm 4.259, time 7.26s, 
epoch 198, batches 18183|18183, train-acc 0.868, loss 0.349, para-norm 5119.517, grad-norm 15.880, time 1.33s, 
dev-acc 0.860
epoch 199, batches 1000|18183, train-acc 0.862, loss 0.358, para-norm 5119.686, grad-norm 5.936, time 7.31s, 
epoch 199, batches 2000|18183, train-acc 0.865, loss 0.350, para-norm 5119.618, grad-norm 2.375, time 7.28s, 
epoch 199, batches 3000|18183, train-acc 0.867, loss 0.348, para-norm 5119.103, grad-norm 10.309, time 7.26s, 
epoch 199, batches 4000|18183, train-acc 0.867, loss 0.353, para-norm 5119.575, grad-norm 1.529, time 7.27s, 
epoch 199, batches 5000|18183, train-acc 0.862, loss 0.361, para-norm 5119.093, grad-norm 5.628, time 7.26s, 
epoch 199, batches 6000|18183, train-acc 0.866, loss 0.356, para-norm 5119.295, grad-norm 8.670, time 7.26s, 
epoch 199, batches 7000|18183, train-acc 0.864, loss 0.353, para-norm 5119.418, grad-norm 6.452, time 7.27s, 
epoch 199, batches 8000|18183, train-acc 0.864, loss 0.355, para-norm 5119.302, grad-norm 6.112, time 7.27s, 
epoch 199, batches 9000|18183, train-acc 0.863, loss 0.362, para-norm 5119.064, grad-norm 11.389, time 7.27s, 
epoch 199, batches 10000|18183, train-acc 0.864, loss 0.354, para-norm 5118.919, grad-norm 3.166, time 7.27s, 
epoch 199, batches 11000|18183, train-acc 0.863, loss 0.361, para-norm 5119.050, grad-norm 11.937, time 7.27s, 
epoch 199, batches 12000|18183, train-acc 0.864, loss 0.365, para-norm 5118.319, grad-norm 5.839, time 7.24s, 
epoch 199, batches 13000|18183, train-acc 0.862, loss 0.366, para-norm 5117.935, grad-norm 2.141, time 7.28s, 
epoch 199, batches 14000|18183, train-acc 0.864, loss 0.356, para-norm 5117.801, grad-norm 2.424, time 7.27s, 
epoch 199, batches 15000|18183, train-acc 0.865, loss 0.354, para-norm 5118.045, grad-norm 13.035, time 7.26s, 
epoch 199, batches 16000|18183, train-acc 0.866, loss 0.352, para-norm 5118.435, grad-norm 9.152, time 7.26s, 
epoch 199, batches 17000|18183, train-acc 0.858, loss 0.371, para-norm 5117.687, grad-norm 12.671, time 7.27s, 
epoch 199, batches 18000|18183, train-acc 0.860, loss 0.369, para-norm 5117.551, grad-norm 9.776, time 7.27s, 
epoch 199, batches 18183|18183, train-acc 0.865, loss 0.359, para-norm 5117.485, grad-norm 2.730, time 1.33s, 
dev-acc 0.862
epoch 200, batches 1000|18183, train-acc 0.863, loss 0.364, para-norm 5117.453, grad-norm 6.335, time 7.34s, 
epoch 200, batches 2000|18183, train-acc 0.864, loss 0.360, para-norm 5117.973, grad-norm 6.252, time 7.27s, 
epoch 200, batches 3000|18183, train-acc 0.863, loss 0.362, para-norm 5118.442, grad-norm 5.826, time 7.27s, 
epoch 200, batches 4000|18183, train-acc 0.865, loss 0.358, para-norm 5118.956, grad-norm 9.746, time 7.28s, 
epoch 200, batches 5000|18183, train-acc 0.864, loss 0.355, para-norm 5118.793, grad-norm 2.002, time 7.26s, 
epoch 200, batches 6000|18183, train-acc 0.866, loss 0.351, para-norm 5118.647, grad-norm 2.366, time 7.27s, 
epoch 200, batches 7000|18183, train-acc 0.865, loss 0.359, para-norm 5118.402, grad-norm 10.623, time 7.27s, 
epoch 200, batches 8000|18183, train-acc 0.866, loss 0.356, para-norm 5118.169, grad-norm 980.585, time 7.27s, 
epoch 200, batches 9000|18183, train-acc 0.862, loss 0.365, para-norm 5117.800, grad-norm 2.786, time 7.29s, 
epoch 200, batches 10000|18183, train-acc 0.863, loss 0.359, para-norm 5118.089, grad-norm 5.424, time 7.28s, 
epoch 200, batches 11000|18183, train-acc 0.866, loss 0.353, para-norm 5118.314, grad-norm 4.710, time 7.27s, 
epoch 200, batches 12000|18183, train-acc 0.864, loss 0.359, para-norm 5117.670, grad-norm 3.614, time 7.27s, 
epoch 200, batches 13000|18183, train-acc 0.863, loss 0.365, para-norm 5117.291, grad-norm 7.786, time 7.28s, 
epoch 200, batches 14000|18183, train-acc 0.864, loss 0.356, para-norm 5116.785, grad-norm 11.355, time 7.27s, 
epoch 200, batches 15000|18183, train-acc 0.865, loss 0.355, para-norm 5116.927, grad-norm 7.423, time 7.27s, 
epoch 200, batches 16000|18183, train-acc 0.866, loss 0.355, para-norm 5116.912, grad-norm 5.644, time 7.27s, 
epoch 200, batches 17000|18183, train-acc 0.865, loss 0.352, para-norm 5116.752, grad-norm 7.323, time 7.27s, 
epoch 200, batches 18000|18183, train-acc 0.863, loss 0.361, para-norm 5116.323, grad-norm 18.339, time 7.28s, 
epoch 200, batches 18183|18183, train-acc 0.867, loss 0.356, para-norm 5116.145, grad-norm 6.880, time 1.33s, 
dev-acc 0.861
epoch 201, batches 1000|18183, train-acc 0.861, loss 0.360, para-norm 5116.095, grad-norm 9.893, time 7.38s, 
epoch 201, batches 2000|18183, train-acc 0.861, loss 0.370, para-norm 5116.106, grad-norm 2.087, time 7.27s, 
epoch 201, batches 3000|18183, train-acc 0.863, loss 0.363, para-norm 5116.178, grad-norm 18.640, time 7.27s, 
epoch 201, batches 4000|18183, train-acc 0.866, loss 0.357, para-norm 5116.063, grad-norm 9.761, time 7.36s, 
epoch 201, batches 5000|18183, train-acc 0.865, loss 0.354, para-norm 5115.942, grad-norm 1.826, time 7.28s, 
epoch 201, batches 6000|18183, train-acc 0.866, loss 0.354, para-norm 5115.970, grad-norm 3.853, time 7.25s, 
epoch 201, batches 7000|18183, train-acc 0.865, loss 0.351, para-norm 5115.944, grad-norm 6.272, time 7.26s, 
epoch 201, batches 8000|18183, train-acc 0.868, loss 0.351, para-norm 5116.002, grad-norm 10.371, time 7.28s, 
epoch 201, batches 9000|18183, train-acc 0.860, loss 0.364, para-norm 5115.724, grad-norm 3.438, time 7.27s, 
epoch 201, batches 10000|18183, train-acc 0.861, loss 0.355, para-norm 5115.623, grad-norm 4.550, time 7.38s, 
epoch 201, batches 11000|18183, train-acc 0.862, loss 0.364, para-norm 5115.339, grad-norm 3.781, time 7.40s, 
epoch 201, batches 12000|18183, train-acc 0.861, loss 0.368, para-norm 5115.160, grad-norm 8.585, time 7.27s, 
epoch 201, batches 13000|18183, train-acc 0.864, loss 0.357, para-norm 5114.898, grad-norm 1.275, time 7.26s, 
epoch 201, batches 14000|18183, train-acc 0.863, loss 0.358, para-norm 5114.881, grad-norm 3.352, time 7.27s, 
epoch 201, batches 15000|18183, train-acc 0.869, loss 0.346, para-norm 5114.525, grad-norm 7.903, time 7.26s, 
epoch 201, batches 16000|18183, train-acc 0.866, loss 0.355, para-norm 5114.152, grad-norm 3.988, time 7.28s, 
epoch 201, batches 17000|18183, train-acc 0.863, loss 0.358, para-norm 5114.317, grad-norm 6.228, time 7.28s, 
epoch 201, batches 18000|18183, train-acc 0.864, loss 0.362, para-norm 5113.841, grad-norm 6.279, time 7.27s, 
epoch 201, batches 18183|18183, train-acc 0.854, loss 0.372, para-norm 5113.842, grad-norm 3.605, time 1.37s, 
dev-acc 0.860
epoch 202, batches 1000|18183, train-acc 0.867, loss 0.350, para-norm 5114.167, grad-norm 3.591, time 7.67s, 
epoch 202, batches 2000|18183, train-acc 0.863, loss 0.361, para-norm 5114.722, grad-norm 2.043, time 7.26s, 
epoch 202, batches 3000|18183, train-acc 0.867, loss 0.352, para-norm 5115.038, grad-norm 4.335, time 7.27s, 
epoch 202, batches 4000|18183, train-acc 0.861, loss 0.365, para-norm 5115.550, grad-norm 247.190, time 7.27s, 
epoch 202, batches 5000|18183, train-acc 0.868, loss 0.352, para-norm 5115.200, grad-norm 6.264, time 7.27s, 
epoch 202, batches 6000|18183, train-acc 0.867, loss 0.356, para-norm 5114.656, grad-norm 4.206, time 7.29s, 
epoch 202, batches 7000|18183, train-acc 0.868, loss 0.348, para-norm 5114.809, grad-norm 4.716, time 7.25s, 
epoch 202, batches 8000|18183, train-acc 0.862, loss 0.361, para-norm 5114.666, grad-norm 4.017, time 7.28s, 
epoch 202, batches 9000|18183, train-acc 0.859, loss 0.368, para-norm 5114.158, grad-norm 6.372, time 7.27s, 
epoch 202, batches 10000|18183, train-acc 0.864, loss 0.357, para-norm 5113.785, grad-norm 5.535, time 7.26s, 
epoch 202, batches 11000|18183, train-acc 0.868, loss 0.348, para-norm 5114.088, grad-norm 5.327, time 7.27s, 
epoch 202, batches 12000|18183, train-acc 0.864, loss 0.360, para-norm 5114.331, grad-norm 4.231, time 7.27s, 
epoch 202, batches 13000|18183, train-acc 0.868, loss 0.351, para-norm 5114.142, grad-norm 7.274, time 7.25s, 
epoch 202, batches 14000|18183, train-acc 0.863, loss 0.360, para-norm 5113.403, grad-norm 5.533, time 7.27s, 
epoch 202, batches 15000|18183, train-acc 0.867, loss 0.355, para-norm 5112.798, grad-norm 9.893, time 7.27s, 
epoch 202, batches 16000|18183, train-acc 0.857, loss 0.369, para-norm 5111.752, grad-norm 6.118, time 7.29s, 
epoch 202, batches 17000|18183, train-acc 0.865, loss 0.357, para-norm 5111.868, grad-norm 752.803, time 7.27s, 
epoch 202, batches 18000|18183, train-acc 0.864, loss 0.359, para-norm 5111.503, grad-norm 8.933, time 7.28s, 
epoch 202, batches 18183|18183, train-acc 0.857, loss 0.366, para-norm 5111.406, grad-norm 0.864, time 1.33s, 
dev-acc 0.862
epoch 203, batches 1000|18183, train-acc 0.864, loss 0.359, para-norm 5117.366, grad-norm 1.193, time 7.34s, 
epoch 203, batches 2000|18183, train-acc 0.864, loss 0.355, para-norm 5118.802, grad-norm 4.555, time 7.27s, 
epoch 203, batches 3000|18183, train-acc 0.864, loss 0.356, para-norm 5118.841, grad-norm 599.733, time 7.26s, 
epoch 203, batches 4000|18183, train-acc 0.866, loss 0.356, para-norm 5118.908, grad-norm 2.540, time 7.27s, 
epoch 203, batches 5000|18183, train-acc 0.865, loss 0.358, para-norm 5118.515, grad-norm 2.325, time 7.27s, 
epoch 203, batches 6000|18183, train-acc 0.865, loss 0.358, para-norm 5118.296, grad-norm 8.156, time 7.28s, 
epoch 203, batches 7000|18183, train-acc 0.866, loss 0.352, para-norm 5118.344, grad-norm 0.800, time 7.27s, 
epoch 203, batches 8000|18183, train-acc 0.864, loss 0.361, para-norm 5117.396, grad-norm 1.855, time 7.28s, 
epoch 203, batches 9000|18183, train-acc 0.864, loss 0.358, para-norm 5117.945, grad-norm 6.353, time 7.26s, 
epoch 203, batches 10000|18183, train-acc 0.863, loss 0.356, para-norm 5117.807, grad-norm 12.863, time 7.27s, 
epoch 203, batches 11000|18183, train-acc 0.865, loss 0.356, para-norm 5117.914, grad-norm 4.023, time 7.27s, 
epoch 203, batches 12000|18183, train-acc 0.866, loss 0.354, para-norm 5117.924, grad-norm 1.892, time 7.26s, 
epoch 203, batches 13000|18183, train-acc 0.866, loss 0.353, para-norm 5117.736, grad-norm 11.854, time 7.26s, 
epoch 203, batches 14000|18183, train-acc 0.864, loss 0.363, para-norm 5116.717, grad-norm 71.429, time 7.40s, 
epoch 203, batches 15000|18183, train-acc 0.863, loss 0.362, para-norm 5116.648, grad-norm 4.674, time 7.27s, 
epoch 203, batches 16000|18183, train-acc 0.860, loss 0.363, para-norm 5116.443, grad-norm 2.129, time 7.27s, 
epoch 203, batches 17000|18183, train-acc 0.864, loss 0.359, para-norm 5116.322, grad-norm 3.543, time 7.27s, 
epoch 203, batches 18000|18183, train-acc 0.865, loss 0.357, para-norm 5116.281, grad-norm 6.895, time 7.27s, 
epoch 203, batches 18183|18183, train-acc 0.860, loss 0.370, para-norm 5116.152, grad-norm 3.989, time 1.33s, 
dev-acc 0.864
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
	142 0.860
	147 0.862
	152 0.862
	185 0.863
	203 0.864
save model!
epoch 204, batches 1000|18183, train-acc 0.866, loss 0.352, para-norm 5115.961, grad-norm 3.052, time 7.44s, 
epoch 204, batches 2000|18183, train-acc 0.865, loss 0.357, para-norm 5115.912, grad-norm 10.248, time 7.27s, 
epoch 204, batches 3000|18183, train-acc 0.863, loss 0.361, para-norm 5116.095, grad-norm 1.309, time 7.29s, 
epoch 204, batches 4000|18183, train-acc 0.860, loss 0.365, para-norm 5115.964, grad-norm 7.402, time 7.28s, 
epoch 204, batches 5000|18183, train-acc 0.866, loss 0.356, para-norm 5115.705, grad-norm 8.949, time 7.26s, 
epoch 204, batches 6000|18183, train-acc 0.864, loss 0.356, para-norm 5116.355, grad-norm 3.709, time 7.27s, 
epoch 204, batches 7000|18183, train-acc 0.866, loss 0.355, para-norm 5116.209, grad-norm 4.707, time 7.28s, 
epoch 204, batches 8000|18183, train-acc 0.867, loss 0.351, para-norm 5116.269, grad-norm 8.359, time 7.25s, 
epoch 204, batches 9000|18183, train-acc 0.864, loss 0.358, para-norm 5115.980, grad-norm 39.870, time 7.27s, 
epoch 204, batches 10000|18183, train-acc 0.867, loss 0.351, para-norm 5115.924, grad-norm 7.447, time 7.28s, 
epoch 204, batches 11000|18183, train-acc 0.863, loss 0.356, para-norm 5115.610, grad-norm 5.266, time 7.28s, 
epoch 204, batches 12000|18183, train-acc 0.861, loss 0.361, para-norm 5115.137, grad-norm 4.204, time 7.28s, 
epoch 204, batches 13000|18183, train-acc 0.864, loss 0.358, para-norm 5114.404, grad-norm 711.709, time 7.26s, 
epoch 204, batches 14000|18183, train-acc 0.864, loss 0.356, para-norm 5114.250, grad-norm 6.523, time 7.28s, 
epoch 204, batches 15000|18183, train-acc 0.864, loss 0.356, para-norm 5113.466, grad-norm 4.725, time 7.26s, 
epoch 204, batches 16000|18183, train-acc 0.866, loss 0.351, para-norm 5114.102, grad-norm 5.661, time 7.28s, 
epoch 204, batches 17000|18183, train-acc 0.862, loss 0.362, para-norm 5113.613, grad-norm 1.506, time 7.28s, 
epoch 204, batches 18000|18183, train-acc 0.862, loss 0.357, para-norm 5113.328, grad-norm 4.114, time 7.27s, 
epoch 204, batches 18183|18183, train-acc 0.870, loss 0.347, para-norm 5113.458, grad-norm 12.130, time 1.32s, 
dev-acc 0.864
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
	142 0.860
	147 0.862
	152 0.862
	185 0.863
	203 0.864
	204 0.864
save model!
epoch 205, batches 1000|18183, train-acc 0.867, loss 0.353, para-norm 5113.453, grad-norm 11.567, time 7.44s, 
epoch 205, batches 2000|18183, train-acc 0.870, loss 0.349, para-norm 5113.845, grad-norm 9.283, time 7.25s, 
epoch 205, batches 3000|18183, train-acc 0.868, loss 0.352, para-norm 5114.232, grad-norm 14.754, time 7.44s, 
epoch 205, batches 4000|18183, train-acc 0.866, loss 0.353, para-norm 5113.901, grad-norm 14.960, time 7.28s, 
epoch 205, batches 5000|18183, train-acc 0.866, loss 0.351, para-norm 5113.360, grad-norm 5.770, time 7.26s, 
epoch 205, batches 6000|18183, train-acc 0.865, loss 0.354, para-norm 5113.685, grad-norm 5.568, time 7.26s, 
epoch 205, batches 7000|18183, train-acc 0.861, loss 0.361, para-norm 5113.475, grad-norm 168.234, time 7.28s, 
epoch 205, batches 8000|18183, train-acc 0.865, loss 0.354, para-norm 5113.531, grad-norm 5.118, time 7.27s, 
epoch 205, batches 9000|18183, train-acc 0.865, loss 0.356, para-norm 5112.838, grad-norm 4.738, time 7.28s, 
epoch 205, batches 10000|18183, train-acc 0.862, loss 0.359, para-norm 5112.833, grad-norm 0.000, time 7.27s, 
epoch 205, batches 11000|18183, train-acc 0.860, loss 0.368, para-norm 5112.049, grad-norm 2.622, time 7.28s, 
epoch 205, batches 12000|18183, train-acc 0.870, loss 0.344, para-norm 5112.185, grad-norm 1.554, time 7.36s, 
epoch 205, batches 13000|18183, train-acc 0.865, loss 0.357, para-norm 5112.347, grad-norm 6.222, time 7.28s, 
epoch 205, batches 14000|18183, train-acc 0.864, loss 0.363, para-norm 5111.989, grad-norm 6.838, time 7.26s, 
epoch 205, batches 15000|18183, train-acc 0.863, loss 0.365, para-norm 5112.425, grad-norm 7.882, time 7.27s, 
epoch 205, batches 16000|18183, train-acc 0.866, loss 0.355, para-norm 5112.744, grad-norm 8.024, time 7.26s, 
epoch 205, batches 17000|18183, train-acc 0.866, loss 0.359, para-norm 5112.517, grad-norm 6.928, time 7.26s, 
epoch 205, batches 18000|18183, train-acc 0.865, loss 0.359, para-norm 5112.406, grad-norm 1.747, time 7.28s, 
epoch 205, batches 18183|18183, train-acc 0.859, loss 0.366, para-norm 5112.556, grad-norm 5.128, time 1.33s, 
dev-acc 0.861
epoch 206, batches 1000|18183, train-acc 0.866, loss 0.354, para-norm 5112.214, grad-norm 5.225, time 7.34s, 
epoch 206, batches 2000|18183, train-acc 0.866, loss 0.350, para-norm 5112.159, grad-norm 8.168, time 7.28s, 
epoch 206, batches 3000|18183, train-acc 0.862, loss 0.356, para-norm 5112.270, grad-norm 2.946, time 7.38s, 
epoch 206, batches 4000|18183, train-acc 0.866, loss 0.356, para-norm 5112.060, grad-norm 5.081, time 7.41s, 
epoch 206, batches 5000|18183, train-acc 0.867, loss 0.351, para-norm 5112.153, grad-norm 2.731, time 7.28s, 
epoch 206, batches 6000|18183, train-acc 0.867, loss 0.350, para-norm 5111.873, grad-norm 5.256, time 7.26s, 
epoch 206, batches 7000|18183, train-acc 0.864, loss 0.362, para-norm 5111.536, grad-norm 1.793, time 7.29s, 
epoch 206, batches 8000|18183, train-acc 0.865, loss 0.356, para-norm 5110.693, grad-norm 9.423, time 7.27s, 
epoch 206, batches 9000|18183, train-acc 0.866, loss 0.352, para-norm 5111.238, grad-norm 7.491, time 7.34s, 
epoch 206, batches 10000|18183, train-acc 0.864, loss 0.356, para-norm 5111.143, grad-norm 4.171, time 7.27s, 
epoch 206, batches 11000|18183, train-acc 0.861, loss 0.360, para-norm 5110.661, grad-norm 5.812, time 7.27s, 
epoch 206, batches 12000|18183, train-acc 0.865, loss 0.352, para-norm 5110.389, grad-norm 8.571, time 7.26s, 
epoch 206, batches 13000|18183, train-acc 0.863, loss 0.366, para-norm 5109.853, grad-norm 4.162, time 7.27s, 
epoch 206, batches 14000|18183, train-acc 0.866, loss 0.356, para-norm 5109.706, grad-norm 1.769, time 7.36s, 
epoch 206, batches 15000|18183, train-acc 0.865, loss 0.358, para-norm 5109.790, grad-norm 17.829, time 7.26s, 
epoch 206, batches 16000|18183, train-acc 0.865, loss 0.356, para-norm 5109.656, grad-norm 4.964, time 7.28s, 
epoch 206, batches 17000|18183, train-acc 0.865, loss 0.354, para-norm 5109.079, grad-norm 4.025, time 7.26s, 
epoch 206, batches 18000|18183, train-acc 0.862, loss 0.361, para-norm 5109.508, grad-norm 4.388, time 7.29s, 
epoch 206, batches 18183|18183, train-acc 0.864, loss 0.360, para-norm 5109.527, grad-norm 6.637, time 1.33s, 
dev-acc 0.861
epoch 207, batches 1000|18183, train-acc 0.866, loss 0.354, para-norm 5109.229, grad-norm 33.804, time 7.30s, 
epoch 207, batches 2000|18183, train-acc 0.867, loss 0.346, para-norm 5109.931, grad-norm 4.080, time 7.27s, 
epoch 207, batches 3000|18183, train-acc 0.862, loss 0.357, para-norm 5109.312, grad-norm 14.244, time 7.27s, 
epoch 207, batches 4000|18183, train-acc 0.865, loss 0.355, para-norm 5109.754, grad-norm 5.921, time 7.28s, 
epoch 207, batches 5000|18183, train-acc 0.868, loss 0.347, para-norm 5109.717, grad-norm 4.897, time 7.26s, 
epoch 207, batches 6000|18183, train-acc 0.868, loss 0.352, para-norm 5109.555, grad-norm 4.022, time 7.37s, 
epoch 207, batches 7000|18183, train-acc 0.864, loss 0.352, para-norm 5108.948, grad-norm 3.668, time 7.36s, 
epoch 207, batches 8000|18183, train-acc 0.866, loss 0.353, para-norm 5109.302, grad-norm 5.420, time 7.27s, 
epoch 207, batches 9000|18183, train-acc 0.867, loss 0.351, para-norm 5109.828, grad-norm 5.776, time 7.28s, 
epoch 207, batches 10000|18183, train-acc 0.863, loss 0.364, para-norm 5109.074, grad-norm 8.981, time 7.28s, 
epoch 207, batches 11000|18183, train-acc 0.864, loss 0.355, para-norm 5108.865, grad-norm 8.984, time 7.28s, 
epoch 207, batches 12000|18183, train-acc 0.862, loss 0.363, para-norm 5108.435, grad-norm 5.146, time 7.25s, 
epoch 207, batches 13000|18183, train-acc 0.865, loss 0.360, para-norm 5108.125, grad-norm 2.837, time 7.28s, 
epoch 207, batches 14000|18183, train-acc 0.865, loss 0.359, para-norm 5108.016, grad-norm 8.427, time 7.26s, 
epoch 207, batches 15000|18183, train-acc 0.863, loss 0.356, para-norm 5107.586, grad-norm 7.048, time 7.27s, 
epoch 207, batches 16000|18183, train-acc 0.865, loss 0.355, para-norm 5107.844, grad-norm 17.252, time 7.26s, 
epoch 207, batches 17000|18183, train-acc 0.863, loss 0.358, para-norm 5107.774, grad-norm 4.398, time 7.27s, 
epoch 207, batches 18000|18183, train-acc 0.861, loss 0.361, para-norm 5108.021, grad-norm 3.406, time 7.28s, 
epoch 207, batches 18183|18183, train-acc 0.861, loss 0.361, para-norm 5108.012, grad-norm 3.122, time 1.33s, 
dev-acc 0.861
epoch 208, batches 1000|18183, train-acc 0.868, loss 0.348, para-norm 5108.312, grad-norm 9.722, time 7.34s, 
epoch 208, batches 2000|18183, train-acc 0.863, loss 0.361, para-norm 5108.220, grad-norm 3.615, time 7.25s, 
epoch 208, batches 3000|18183, train-acc 0.868, loss 0.349, para-norm 5111.038, grad-norm 1.414, time 7.26s, 
epoch 208, batches 4000|18183, train-acc 0.865, loss 0.358, para-norm 5111.114, grad-norm 2.297, time 7.26s, 
epoch 208, batches 5000|18183, train-acc 0.869, loss 0.347, para-norm 5111.816, grad-norm 1104.826, time 7.26s, 
epoch 208, batches 6000|18183, train-acc 0.863, loss 0.364, para-norm 5110.812, grad-norm 29.933, time 7.28s, 
epoch 208, batches 7000|18183, train-acc 0.867, loss 0.354, para-norm 5111.313, grad-norm 1.391, time 7.27s, 
epoch 208, batches 8000|18183, train-acc 0.863, loss 0.362, para-norm 5111.183, grad-norm 6.724, time 7.26s, 
epoch 208, batches 9000|18183, train-acc 0.867, loss 0.354, para-norm 5111.049, grad-norm 7.972, time 7.26s, 
epoch 208, batches 10000|18183, train-acc 0.862, loss 0.358, para-norm 5111.108, grad-norm 9.856, time 7.28s, 
epoch 208, batches 11000|18183, train-acc 0.864, loss 0.357, para-norm 5110.224, grad-norm 5.297, time 7.28s, 
epoch 208, batches 12000|18183, train-acc 0.867, loss 0.349, para-norm 5110.661, grad-norm 8.077, time 7.27s, 
epoch 208, batches 13000|18183, train-acc 0.862, loss 0.361, para-norm 5110.240, grad-norm 17.677, time 7.27s, 
epoch 208, batches 14000|18183, train-acc 0.866, loss 0.355, para-norm 5109.955, grad-norm 7.553, time 7.27s, 
epoch 208, batches 15000|18183, train-acc 0.863, loss 0.360, para-norm 5109.832, grad-norm 12.317, time 7.26s, 
epoch 208, batches 16000|18183, train-acc 0.865, loss 0.355, para-norm 5110.810, grad-norm 6.961, time 7.26s, 
epoch 208, batches 17000|18183, train-acc 0.862, loss 0.359, para-norm 5110.163, grad-norm 8.178, time 7.27s, 
epoch 208, batches 18000|18183, train-acc 0.862, loss 0.361, para-norm 5109.530, grad-norm 11.986, time 7.26s, 
epoch 208, batches 18183|18183, train-acc 0.871, loss 0.344, para-norm 5109.863, grad-norm 2.455, time 1.33s, 
dev-acc 0.861
epoch 209, batches 1000|18183, train-acc 0.866, loss 0.355, para-norm 5109.475, grad-norm 4.535, time 7.35s, 
epoch 209, batches 2000|18183, train-acc 0.864, loss 0.357, para-norm 5109.461, grad-norm 3.873, time 7.27s, 
epoch 209, batches 3000|18183, train-acc 0.865, loss 0.351, para-norm 5109.582, grad-norm 7.062, time 7.25s, 
epoch 209, batches 4000|18183, train-acc 0.863, loss 0.361, para-norm 5109.441, grad-norm 1.742, time 7.27s, 
epoch 209, batches 5000|18183, train-acc 0.866, loss 0.348, para-norm 5109.344, grad-norm 6.979, time 7.29s, 
epoch 209, batches 6000|18183, train-acc 0.865, loss 0.351, para-norm 5109.193, grad-norm 9.817, time 7.27s, 
epoch 209, batches 7000|18183, train-acc 0.867, loss 0.349, para-norm 5108.915, grad-norm 1.295, time 7.26s, 
epoch 209, batches 8000|18183, train-acc 0.863, loss 0.357, para-norm 5108.390, grad-norm 7.972, time 7.35s, 
epoch 209, batches 9000|18183, train-acc 0.867, loss 0.349, para-norm 5108.465, grad-norm 3.132, time 7.40s, 
epoch 209, batches 10000|18183, train-acc 0.861, loss 0.366, para-norm 5107.775, grad-norm 2.382, time 7.29s, 
epoch 209, batches 11000|18183, train-acc 0.862, loss 0.363, para-norm 5107.952, grad-norm 10.873, time 7.37s, 
epoch 209, batches 12000|18183, train-acc 0.866, loss 0.353, para-norm 5107.392, grad-norm 6.336, time 7.38s, 
epoch 209, batches 13000|18183, train-acc 0.864, loss 0.358, para-norm 5107.131, grad-norm 5.867, time 7.27s, 
epoch 209, batches 14000|18183, train-acc 0.864, loss 0.358, para-norm 5106.354, grad-norm 6.914, time 7.26s, 
epoch 209, batches 15000|18183, train-acc 0.866, loss 0.350, para-norm 5106.411, grad-norm 7.049, time 7.25s, 
epoch 209, batches 16000|18183, train-acc 0.862, loss 0.364, para-norm 5106.203, grad-norm 5.307, time 7.27s, 
epoch 209, batches 17000|18183, train-acc 0.867, loss 0.357, para-norm 5106.625, grad-norm 3.962, time 7.31s, 
epoch 209, batches 18000|18183, train-acc 0.865, loss 0.355, para-norm 5106.307, grad-norm 449.256, time 7.29s, 
epoch 209, batches 18183|18183, train-acc 0.860, loss 0.365, para-norm 5106.384, grad-norm 4.360, time 1.33s, 
dev-acc 0.862
epoch 210, batches 1000|18183, train-acc 0.864, loss 0.356, para-norm 5106.114, grad-norm 3.827, time 7.32s, 
epoch 210, batches 2000|18183, train-acc 0.867, loss 0.347, para-norm 5106.645, grad-norm 6.267, time 7.27s, 
epoch 210, batches 3000|18183, train-acc 0.867, loss 0.355, para-norm 5106.460, grad-norm 2.459, time 7.27s, 
epoch 210, batches 4000|18183, train-acc 0.865, loss 0.358, para-norm 5106.638, grad-norm 1.324, time 7.28s, 
epoch 210, batches 5000|18183, train-acc 0.865, loss 0.351, para-norm 5106.536, grad-norm 7.031, time 7.26s, 
epoch 210, batches 6000|18183, train-acc 0.867, loss 0.353, para-norm 5106.122, grad-norm 6.726, time 7.26s, 
epoch 210, batches 7000|18183, train-acc 0.862, loss 0.362, para-norm 5105.664, grad-norm 9.677, time 7.28s, 
epoch 210, batches 8000|18183, train-acc 0.867, loss 0.354, para-norm 5105.777, grad-norm 2.926, time 7.27s, 
epoch 210, batches 9000|18183, train-acc 0.867, loss 0.350, para-norm 5105.685, grad-norm 6.648, time 7.36s, 
epoch 210, batches 10000|18183, train-acc 0.865, loss 0.354, para-norm 5105.226, grad-norm 3.214, time 7.28s, 
epoch 210, batches 11000|18183, train-acc 0.865, loss 0.351, para-norm 5104.602, grad-norm 2.962, time 7.28s, 
epoch 210, batches 12000|18183, train-acc 0.862, loss 0.363, para-norm 5104.930, grad-norm 4.305, time 7.28s, 
epoch 210, batches 13000|18183, train-acc 0.862, loss 0.359, para-norm 5104.867, grad-norm 5.548, time 7.29s, 
epoch 210, batches 14000|18183, train-acc 0.864, loss 0.357, para-norm 5104.494, grad-norm 2.816, time 7.38s, 
epoch 210, batches 15000|18183, train-acc 0.865, loss 0.357, para-norm 5104.425, grad-norm 7.148, time 7.29s, 
epoch 210, batches 16000|18183, train-acc 0.868, loss 0.350, para-norm 5104.604, grad-norm 3.604, time 7.25s, 
epoch 210, batches 17000|18183, train-acc 0.859, loss 0.368, para-norm 5104.078, grad-norm 7.796, time 7.28s, 
epoch 210, batches 18000|18183, train-acc 0.866, loss 0.351, para-norm 5104.082, grad-norm 2.873, time 7.26s, 
epoch 210, batches 18183|18183, train-acc 0.860, loss 0.365, para-norm 5103.791, grad-norm 5.754, time 1.33s, 
dev-acc 0.862
epoch 211, batches 1000|18183, train-acc 0.867, loss 0.353, para-norm 5103.846, grad-norm 4.558, time 7.35s, 
epoch 211, batches 2000|18183, train-acc 0.864, loss 0.358, para-norm 5103.776, grad-norm 7.499, time 7.28s, 
epoch 211, batches 3000|18183, train-acc 0.863, loss 0.357, para-norm 5103.168, grad-norm 9.859, time 7.28s, 
epoch 211, batches 4000|18183, train-acc 0.865, loss 0.354, para-norm 5103.398, grad-norm 3.432, time 7.27s, 
epoch 211, batches 5000|18183, train-acc 0.868, loss 0.352, para-norm 5103.605, grad-norm 3.753, time 7.28s, 
epoch 211, batches 6000|18183, train-acc 0.868, loss 0.352, para-norm 5103.151, grad-norm 7.583, time 7.27s, 
epoch 211, batches 7000|18183, train-acc 0.863, loss 0.358, para-norm 5102.773, grad-norm 3.113, time 7.46s, 
epoch 211, batches 8000|18183, train-acc 0.868, loss 0.347, para-norm 5103.357, grad-norm 4.558, time 7.61s, 
epoch 211, batches 9000|18183, train-acc 0.865, loss 0.358, para-norm 5103.101, grad-norm 4.699, time 7.62s, 
epoch 211, batches 10000|18183, train-acc 0.864, loss 0.359, para-norm 5102.812, grad-norm 14.412, time 7.63s, 
epoch 211, batches 11000|18183, train-acc 0.865, loss 0.359, para-norm 5102.278, grad-norm 2.863, time 7.61s, 
epoch 211, batches 12000|18183, train-acc 0.862, loss 0.361, para-norm 5101.432, grad-norm 5.413, time 7.61s, 
epoch 211, batches 13000|18183, train-acc 0.869, loss 0.346, para-norm 5101.312, grad-norm 11.784, time 7.61s, 
epoch 211, batches 14000|18183, train-acc 0.869, loss 0.346, para-norm 5105.609, grad-norm 2.439, time 7.61s, 
epoch 211, batches 15000|18183, train-acc 0.866, loss 0.357, para-norm 5105.580, grad-norm 4.984, time 7.61s, 
epoch 211, batches 16000|18183, train-acc 0.862, loss 0.361, para-norm 5104.859, grad-norm 4.802, time 7.61s, 
epoch 211, batches 17000|18183, train-acc 0.866, loss 0.353, para-norm 5104.582, grad-norm 4.871, time 7.61s, 
epoch 211, batches 18000|18183, train-acc 0.867, loss 0.355, para-norm 5104.652, grad-norm 9.460, time 7.60s, 
epoch 211, batches 18183|18183, train-acc 0.861, loss 0.366, para-norm 5104.283, grad-norm 8.472, time 1.39s, 
dev-acc 0.863
epoch 212, batches 1000|18183, train-acc 0.865, loss 0.354, para-norm 5104.615, grad-norm 3.013, time 7.67s, 
epoch 212, batches 2000|18183, train-acc 0.865, loss 0.352, para-norm 5105.118, grad-norm 7.186, time 7.61s, 
epoch 212, batches 3000|18183, train-acc 0.867, loss 0.344, para-norm 5104.867, grad-norm 2.911, time 7.60s, 
epoch 212, batches 4000|18183, train-acc 0.870, loss 0.346, para-norm 5105.772, grad-norm 8.177, time 7.62s, 
epoch 212, batches 5000|18183, train-acc 0.864, loss 0.355, para-norm 5105.276, grad-norm 2.729, time 7.62s, 
epoch 212, batches 6000|18183, train-acc 0.863, loss 0.362, para-norm 5105.088, grad-norm 2.280, time 7.62s, 
epoch 212, batches 7000|18183, train-acc 0.864, loss 0.357, para-norm 5104.879, grad-norm 4.362, time 7.63s, 
epoch 212, batches 8000|18183, train-acc 0.860, loss 0.359, para-norm 5104.912, grad-norm 5.095, time 7.62s, 
epoch 212, batches 9000|18183, train-acc 0.864, loss 0.356, para-norm 5104.940, grad-norm 105.621, time 7.62s, 
epoch 212, batches 10000|18183, train-acc 0.865, loss 0.356, para-norm 5104.992, grad-norm 3.584, time 7.62s, 
epoch 212, batches 11000|18183, train-acc 0.863, loss 0.356, para-norm 5103.965, grad-norm 4.874, time 7.62s, 
epoch 212, batches 12000|18183, train-acc 0.866, loss 0.348, para-norm 5103.815, grad-norm 27.233, time 7.61s, 
epoch 212, batches 13000|18183, train-acc 0.862, loss 0.360, para-norm 5103.829, grad-norm 3.978, time 7.62s, 
epoch 212, batches 14000|18183, train-acc 0.867, loss 0.353, para-norm 5103.545, grad-norm 3.204, time 7.60s, 
epoch 212, batches 15000|18183, train-acc 0.866, loss 0.356, para-norm 5103.381, grad-norm 6.514, time 7.61s, 
epoch 212, batches 16000|18183, train-acc 0.867, loss 0.356, para-norm 5103.403, grad-norm 2.052, time 7.61s, 
epoch 212, batches 17000|18183, train-acc 0.863, loss 0.362, para-norm 5102.857, grad-norm 6.948, time 7.62s, 
epoch 212, batches 18000|18183, train-acc 0.865, loss 0.360, para-norm 5101.536, grad-norm 6.968, time 7.62s, 
epoch 212, batches 18183|18183, train-acc 0.868, loss 0.354, para-norm 5101.581, grad-norm 7.352, time 1.40s, 
dev-acc 0.862
epoch 213, batches 1000|18183, train-acc 0.867, loss 0.351, para-norm 5101.712, grad-norm 6.446, time 7.66s, 
epoch 213, batches 2000|18183, train-acc 0.868, loss 0.352, para-norm 5101.815, grad-norm 5.780, time 7.63s, 
epoch 213, batches 3000|18183, train-acc 0.866, loss 0.356, para-norm 5107.085, grad-norm 9.102, time 7.62s, 
epoch 213, batches 4000|18183, train-acc 0.868, loss 0.355, para-norm 5106.476, grad-norm 5.922, time 7.62s, 
epoch 213, batches 5000|18183, train-acc 0.863, loss 0.358, para-norm 5107.114, grad-norm 4.443, time 7.61s, 
epoch 213, batches 6000|18183, train-acc 0.863, loss 0.357, para-norm 5107.220, grad-norm 6.549, time 7.61s, 
epoch 213, batches 7000|18183, train-acc 0.867, loss 0.352, para-norm 5107.258, grad-norm 9.364, time 7.63s, 
epoch 213, batches 8000|18183, train-acc 0.865, loss 0.358, para-norm 5107.108, grad-norm 98.579, time 7.61s, 
epoch 213, batches 9000|18183, train-acc 0.865, loss 0.356, para-norm 5106.976, grad-norm 14.488, time 7.62s, 
epoch 213, batches 10000|18183, train-acc 0.867, loss 0.351, para-norm 5107.072, grad-norm 5.630, time 7.61s, 
epoch 213, batches 11000|18183, train-acc 0.867, loss 0.353, para-norm 5106.450, grad-norm 6.672, time 7.62s, 
epoch 213, batches 12000|18183, train-acc 0.866, loss 0.355, para-norm 5106.104, grad-norm 8.176, time 7.61s, 
epoch 213, batches 13000|18183, train-acc 0.862, loss 0.363, para-norm 5105.714, grad-norm 3.482, time 7.62s, 
epoch 213, batches 14000|18183, train-acc 0.865, loss 0.355, para-norm 5105.240, grad-norm 2.678, time 7.61s, 
epoch 213, batches 15000|18183, train-acc 0.863, loss 0.359, para-norm 5105.484, grad-norm 389.799, time 7.63s, 
epoch 213, batches 16000|18183, train-acc 0.862, loss 0.360, para-norm 5104.865, grad-norm 5.305, time 7.61s, 
epoch 213, batches 17000|18183, train-acc 0.868, loss 0.346, para-norm 5104.893, grad-norm 6.426, time 7.60s, 
epoch 213, batches 18000|18183, train-acc 0.858, loss 0.369, para-norm 5110.049, grad-norm 11.996, time 7.63s, 
epoch 213, batches 18183|18183, train-acc 0.862, loss 0.360, para-norm 5110.183, grad-norm 3.094, time 1.39s, 
dev-acc 0.861
epoch 214, batches 1000|18183, train-acc 0.871, loss 0.343, para-norm 5111.058, grad-norm 3.495, time 7.67s, 
epoch 214, batches 2000|18183, train-acc 0.867, loss 0.349, para-norm 5109.626, grad-norm 3.729, time 7.61s, 
epoch 214, batches 3000|18183, train-acc 0.869, loss 0.345, para-norm 5109.300, grad-norm 4.150, time 7.62s, 
epoch 214, batches 4000|18183, train-acc 0.866, loss 0.356, para-norm 5109.133, grad-norm 2.759, time 7.62s, 
epoch 214, batches 5000|18183, train-acc 0.868, loss 0.352, para-norm 5107.965, grad-norm 12.075, time 7.62s, 
epoch 214, batches 6000|18183, train-acc 0.867, loss 0.353, para-norm 5108.288, grad-norm 2.491, time 7.61s, 
epoch 214, batches 7000|18183, train-acc 0.865, loss 0.355, para-norm 5108.183, grad-norm 7.640, time 7.63s, 
epoch 214, batches 8000|18183, train-acc 0.867, loss 0.351, para-norm 5107.860, grad-norm 4.269, time 7.62s, 
epoch 214, batches 9000|18183, train-acc 0.864, loss 0.358, para-norm 5107.690, grad-norm 6.105, time 7.61s, 
epoch 214, batches 10000|18183, train-acc 0.863, loss 0.360, para-norm 5107.292, grad-norm 1.451, time 7.64s, 
epoch 214, batches 11000|18183, train-acc 0.862, loss 0.359, para-norm 5107.048, grad-norm 11.942, time 7.64s, 
epoch 214, batches 12000|18183, train-acc 0.863, loss 0.361, para-norm 5106.087, grad-norm 3.668, time 7.68s, 
epoch 214, batches 13000|18183, train-acc 0.864, loss 0.358, para-norm 5105.380, grad-norm 4.091, time 7.63s, 
epoch 214, batches 14000|18183, train-acc 0.863, loss 0.360, para-norm 5104.545, grad-norm 5.592, time 7.63s, 
epoch 214, batches 15000|18183, train-acc 0.867, loss 0.353, para-norm 5104.569, grad-norm 0.049, time 7.61s, 
epoch 214, batches 16000|18183, train-acc 0.865, loss 0.358, para-norm 5104.513, grad-norm 14.603, time 7.63s, 
epoch 214, batches 17000|18183, train-acc 0.866, loss 0.352, para-norm 5104.459, grad-norm 6.232, time 7.39s, 
epoch 214, batches 18000|18183, train-acc 0.866, loss 0.360, para-norm 5104.052, grad-norm 3.765, time 7.28s, 
epoch 214, batches 18183|18183, train-acc 0.866, loss 0.351, para-norm 5104.095, grad-norm 5.314, time 1.33s, 
dev-acc 0.863
epoch 215, batches 1000|18183, train-acc 0.864, loss 0.357, para-norm 5104.239, grad-norm 6.528, time 7.35s, 
epoch 215, batches 2000|18183, train-acc 0.865, loss 0.353, para-norm 5104.518, grad-norm 6.574, time 7.30s, 
epoch 215, batches 3000|18183, train-acc 0.866, loss 0.353, para-norm 5104.365, grad-norm 3.483, time 7.12s, 
epoch 215, batches 4000|18183, train-acc 0.872, loss 0.339, para-norm 5104.129, grad-norm 1.085, time 7.10s, 
epoch 215, batches 5000|18183, train-acc 0.868, loss 0.351, para-norm 5103.241, grad-norm 4.013, time 7.57s, 
epoch 215, batches 6000|18183, train-acc 0.865, loss 0.356, para-norm 5102.221, grad-norm 5.262, time 7.49s, 
epoch 215, batches 7000|18183, train-acc 0.866, loss 0.352, para-norm 5102.419, grad-norm 3.544, time 7.48s, 
epoch 215, batches 8000|18183, train-acc 0.868, loss 0.350, para-norm 5102.632, grad-norm 7.366, time 7.51s, 
epoch 215, batches 9000|18183, train-acc 0.869, loss 0.351, para-norm 5102.465, grad-norm 7.522, time 7.55s, 
epoch 215, batches 10000|18183, train-acc 0.868, loss 0.347, para-norm 5102.673, grad-norm 6.614, time 7.58s, 
epoch 215, batches 11000|18183, train-acc 0.867, loss 0.356, para-norm 5102.039, grad-norm 1.057, time 7.37s, 
epoch 215, batches 12000|18183, train-acc 0.864, loss 0.359, para-norm 5101.919, grad-norm 5.067, time 7.56s, 
epoch 215, batches 13000|18183, train-acc 0.866, loss 0.352, para-norm 5101.747, grad-norm 6.003, time 7.55s, 
epoch 215, batches 14000|18183, train-acc 0.862, loss 0.360, para-norm 5101.113, grad-norm 4.457, time 7.51s, 
epoch 215, batches 15000|18183, train-acc 0.867, loss 0.353, para-norm 5101.444, grad-norm 6.877, time 7.54s, 
epoch 215, batches 16000|18183, train-acc 0.869, loss 0.349, para-norm 5101.469, grad-norm 6.618, time 7.26s, 
epoch 215, batches 17000|18183, train-acc 0.864, loss 0.360, para-norm 5101.026, grad-norm 4.793, time 7.13s, 
epoch 215, batches 18000|18183, train-acc 0.863, loss 0.358, para-norm 5100.737, grad-norm 4.053, time 7.13s, 
epoch 215, batches 18183|18183, train-acc 0.853, loss 0.365, para-norm 5100.594, grad-norm 5.183, time 1.31s, 
dev-acc 0.861
epoch 216, batches 1000|18183, train-acc 0.867, loss 0.348, para-norm 5100.171, grad-norm 10.209, time 7.18s, 
epoch 216, batches 2000|18183, train-acc 0.866, loss 0.354, para-norm 5100.226, grad-norm 9.604, time 7.13s, 
epoch 216, batches 3000|18183, train-acc 0.870, loss 0.340, para-norm 5100.167, grad-norm 3.130, time 7.11s, 
epoch 216, batches 4000|18183, train-acc 0.870, loss 0.344, para-norm 5100.105, grad-norm 7.157, time 7.11s, 
epoch 216, batches 5000|18183, train-acc 0.862, loss 0.356, para-norm 5099.655, grad-norm 5.763, time 7.12s, 
epoch 216, batches 6000|18183, train-acc 0.869, loss 0.350, para-norm 5099.526, grad-norm 5.262, time 7.12s, 
epoch 216, batches 7000|18183, train-acc 0.868, loss 0.351, para-norm 5099.765, grad-norm 5.413, time 7.14s, 
epoch 216, batches 8000|18183, train-acc 0.866, loss 0.359, para-norm 5099.559, grad-norm 5.013, time 7.27s, 
epoch 216, batches 9000|18183, train-acc 0.865, loss 0.354, para-norm 5099.621, grad-norm 7.236, time 7.29s, 
epoch 216, batches 10000|18183, train-acc 0.865, loss 0.356, para-norm 5099.198, grad-norm 2.679, time 7.29s, 
epoch 216, batches 11000|18183, train-acc 0.864, loss 0.360, para-norm 5099.448, grad-norm 2.551, time 7.28s, 
epoch 216, batches 12000|18183, train-acc 0.861, loss 0.363, para-norm 5099.476, grad-norm 4.764, time 7.31s, 
epoch 216, batches 13000|18183, train-acc 0.869, loss 0.349, para-norm 5098.446, grad-norm 7.698, time 7.28s, 
epoch 216, batches 14000|18183, train-acc 0.863, loss 0.358, para-norm 5098.568, grad-norm 6.101, time 7.29s, 
epoch 216, batches 15000|18183, train-acc 0.865, loss 0.353, para-norm 5098.443, grad-norm 3.345, time 7.28s, 
epoch 216, batches 16000|18183, train-acc 0.864, loss 0.356, para-norm 5098.279, grad-norm 8.508, time 7.28s, 
epoch 216, batches 17000|18183, train-acc 0.865, loss 0.358, para-norm 5097.742, grad-norm 6.324, time 7.29s, 
epoch 216, batches 18000|18183, train-acc 0.864, loss 0.356, para-norm 5097.268, grad-norm 1.272, time 7.28s, 
epoch 216, batches 18183|18183, train-acc 0.865, loss 0.358, para-norm 5097.013, grad-norm 1.031, time 1.33s, 
dev-acc 0.862
epoch 217, batches 1000|18183, train-acc 0.868, loss 0.348, para-norm 5096.977, grad-norm 4.342, time 7.34s, 
epoch 217, batches 2000|18183, train-acc 0.867, loss 0.347, para-norm 5097.055, grad-norm 9.766, time 7.28s, 
epoch 217, batches 3000|18183, train-acc 0.867, loss 0.352, para-norm 5096.757, grad-norm 4.920, time 7.26s, 
epoch 217, batches 4000|18183, train-acc 0.866, loss 0.357, para-norm 5096.775, grad-norm 2.956, time 7.26s, 
epoch 217, batches 5000|18183, train-acc 0.871, loss 0.342, para-norm 5097.232, grad-norm 11.051, time 7.26s, 
epoch 217, batches 6000|18183, train-acc 0.864, loss 0.363, para-norm 5096.905, grad-norm 16.988, time 7.29s, 
epoch 217, batches 7000|18183, train-acc 0.867, loss 0.351, para-norm 5096.515, grad-norm 5.098, time 7.28s, 
epoch 217, batches 8000|18183, train-acc 0.868, loss 0.353, para-norm 5100.026, grad-norm 8.638, time 7.27s, 
epoch 217, batches 9000|18183, train-acc 0.867, loss 0.356, para-norm 5102.411, grad-norm 6.712, time 7.28s, 
epoch 217, batches 10000|18183, train-acc 0.868, loss 0.348, para-norm 5102.358, grad-norm 5.139, time 7.29s, 
epoch 217, batches 11000|18183, train-acc 0.864, loss 0.357, para-norm 5101.414, grad-norm 8.542, time 7.29s, 
epoch 217, batches 12000|18183, train-acc 0.867, loss 0.354, para-norm 5100.798, grad-norm 10.799, time 7.28s, 
epoch 217, batches 13000|18183, train-acc 0.865, loss 0.354, para-norm 5100.493, grad-norm 7.112, time 7.28s, 
epoch 217, batches 14000|18183, train-acc 0.863, loss 0.356, para-norm 5103.126, grad-norm 5.183, time 7.28s, 
epoch 217, batches 15000|18183, train-acc 0.865, loss 0.351, para-norm 5102.657, grad-norm 6.978, time 7.29s, 
epoch 217, batches 16000|18183, train-acc 0.863, loss 0.359, para-norm 5102.162, grad-norm 2.086, time 7.29s, 
epoch 217, batches 17000|18183, train-acc 0.864, loss 0.360, para-norm 5101.690, grad-norm 11.201, time 7.28s, 
epoch 217, batches 18000|18183, train-acc 0.865, loss 0.355, para-norm 5101.300, grad-norm 5.335, time 7.30s, 
epoch 217, batches 18183|18183, train-acc 0.867, loss 0.346, para-norm 5101.160, grad-norm 5.312, time 1.33s, 
dev-acc 0.864
epoch 218, batches 1000|18183, train-acc 0.868, loss 0.355, para-norm 5100.374, grad-norm 9.942, time 7.32s, 
epoch 218, batches 2000|18183, train-acc 0.866, loss 0.353, para-norm 5100.480, grad-norm 6.723, time 7.28s, 
epoch 218, batches 3000|18183, train-acc 0.866, loss 0.354, para-norm 5100.108, grad-norm 3.848, time 7.28s, 
epoch 218, batches 4000|18183, train-acc 0.865, loss 0.353, para-norm 5106.079, grad-norm 3.445, time 7.28s, 
epoch 218, batches 5000|18183, train-acc 0.867, loss 0.354, para-norm 5105.553, grad-norm 8.512, time 7.28s, 
epoch 218, batches 6000|18183, train-acc 0.869, loss 0.347, para-norm 5105.592, grad-norm 6.831, time 7.27s, 
epoch 218, batches 7000|18183, train-acc 0.864, loss 0.355, para-norm 5105.161, grad-norm 3.203, time 7.29s, 
epoch 218, batches 8000|18183, train-acc 0.865, loss 0.352, para-norm 5105.261, grad-norm 6.332, time 7.28s, 
epoch 218, batches 9000|18183, train-acc 0.866, loss 0.350, para-norm 5104.980, grad-norm 0.002, time 7.28s, 
epoch 218, batches 10000|18183, train-acc 0.868, loss 0.349, para-norm 5105.163, grad-norm 7.294, time 7.28s, 
epoch 218, batches 11000|18183, train-acc 0.868, loss 0.349, para-norm 5111.031, grad-norm 5.175, time 7.28s, 
epoch 218, batches 12000|18183, train-acc 0.865, loss 0.357, para-norm 5110.603, grad-norm 5.426, time 7.29s, 
epoch 218, batches 13000|18183, train-acc 0.865, loss 0.354, para-norm 5109.669, grad-norm 9.640, time 7.42s, 
epoch 218, batches 14000|18183, train-acc 0.867, loss 0.354, para-norm 5109.853, grad-norm 5.584, time 7.29s, 
epoch 218, batches 15000|18183, train-acc 0.862, loss 0.358, para-norm 5109.082, grad-norm 3.405, time 7.28s, 
epoch 218, batches 16000|18183, train-acc 0.867, loss 0.351, para-norm 5109.082, grad-norm 5.904, time 7.27s, 
epoch 218, batches 17000|18183, train-acc 0.867, loss 0.357, para-norm 5108.971, grad-norm 10.793, time 7.28s, 
epoch 218, batches 18000|18183, train-acc 0.863, loss 0.362, para-norm 5108.052, grad-norm 139.966, time 7.28s, 
epoch 218, batches 18183|18183, train-acc 0.869, loss 0.351, para-norm 5108.079, grad-norm 5.358, time 1.33s, 
dev-acc 0.864
epoch 219, batches 1000|18183, train-acc 0.868, loss 0.352, para-norm 5108.383, grad-norm 5.923, time 7.33s, 
epoch 219, batches 2000|18183, train-acc 0.867, loss 0.352, para-norm 5108.107, grad-norm 4.174, time 7.28s, 
epoch 219, batches 3000|18183, train-acc 0.866, loss 0.351, para-norm 5107.662, grad-norm 4.027, time 7.27s, 
epoch 219, batches 4000|18183, train-acc 0.866, loss 0.360, para-norm 5106.620, grad-norm 23.515, time 7.28s, 
epoch 219, batches 5000|18183, train-acc 0.870, loss 0.349, para-norm 5106.568, grad-norm 4.702, time 7.29s, 
epoch 219, batches 6000|18183, train-acc 0.869, loss 0.350, para-norm 5106.372, grad-norm 10.919, time 7.31s, 
epoch 219, batches 7000|18183, train-acc 0.867, loss 0.349, para-norm 5106.678, grad-norm 9.488, time 7.28s, 
epoch 219, batches 8000|18183, train-acc 0.865, loss 0.355, para-norm 5105.735, grad-norm 9.504, time 7.29s, 
epoch 219, batches 9000|18183, train-acc 0.865, loss 0.357, para-norm 5105.043, grad-norm 4.340, time 7.28s, 
epoch 219, batches 10000|18183, train-acc 0.866, loss 0.351, para-norm 5104.965, grad-norm 6.501, time 7.28s, 
epoch 219, batches 11000|18183, train-acc 0.863, loss 0.358, para-norm 5104.627, grad-norm 3.745, time 7.25s, 
epoch 219, batches 12000|18183, train-acc 0.863, loss 0.358, para-norm 5104.196, grad-norm 6.935, time 7.21s, 
epoch 219, batches 13000|18183, train-acc 0.867, loss 0.357, para-norm 5104.018, grad-norm 4.151, time 7.30s, 
epoch 219, batches 14000|18183, train-acc 0.869, loss 0.354, para-norm 5103.422, grad-norm 5.048, time 7.17s, 
epoch 219, batches 15000|18183, train-acc 0.867, loss 0.351, para-norm 5103.595, grad-norm 3.120, time 7.11s, 
epoch 219, batches 16000|18183, train-acc 0.868, loss 0.353, para-norm 5103.392, grad-norm 4.624, time 7.10s, 
epoch 219, batches 17000|18183, train-acc 0.863, loss 0.359, para-norm 5102.861, grad-norm 2.297, time 7.12s, 
epoch 219, batches 18000|18183, train-acc 0.869, loss 0.351, para-norm 5102.004, grad-norm 4.100, time 7.11s, 
epoch 219, batches 18183|18183, train-acc 0.862, loss 0.353, para-norm 5101.774, grad-norm 2.522, time 1.31s, 
dev-acc 0.861
epoch 220, batches 1000|18183, train-acc 0.868, loss 0.345, para-norm 5102.272, grad-norm 3.489, time 7.19s, 
epoch 220, batches 2000|18183, train-acc 0.863, loss 0.359, para-norm 5102.137, grad-norm 4.732, time 7.13s, 
epoch 220, batches 3000|18183, train-acc 0.868, loss 0.348, para-norm 5101.939, grad-norm 9.237, time 7.11s, 
epoch 220, batches 4000|18183, train-acc 0.867, loss 0.348, para-norm 5101.635, grad-norm 3.239, time 7.11s, 
epoch 220, batches 5000|18183, train-acc 0.863, loss 0.359, para-norm 5101.539, grad-norm 7.797, time 7.61s, 
epoch 220, batches 6000|18183, train-acc 0.866, loss 0.351, para-norm 5101.684, grad-norm 5.404, time 7.62s, 
epoch 220, batches 7000|18183, train-acc 0.863, loss 0.358, para-norm 5101.174, grad-norm 2.997, time 7.62s, 
epoch 220, batches 8000|18183, train-acc 0.865, loss 0.359, para-norm 5101.336, grad-norm 10.746, time 7.62s, 
epoch 220, batches 9000|18183, train-acc 0.866, loss 0.355, para-norm 5101.257, grad-norm 2.886, time 7.61s, 
epoch 220, batches 10000|18183, train-acc 0.871, loss 0.344, para-norm 5101.233, grad-norm 5.470, time 7.61s, 
epoch 220, batches 11000|18183, train-acc 0.867, loss 0.352, para-norm 5101.053, grad-norm 6.856, time 7.63s, 
epoch 220, batches 12000|18183, train-acc 0.865, loss 0.357, para-norm 5100.640, grad-norm 3.709, time 7.48s, 
epoch 220, batches 13000|18183, train-acc 0.868, loss 0.348, para-norm 5100.789, grad-norm 14.165, time 7.11s, 
epoch 220, batches 14000|18183, train-acc 0.867, loss 0.354, para-norm 5100.762, grad-norm 3.822, time 7.12s, 
epoch 220, batches 15000|18183, train-acc 0.865, loss 0.354, para-norm 5100.118, grad-norm 7.301, time 7.11s, 
epoch 220, batches 16000|18183, train-acc 0.867, loss 0.357, para-norm 5099.808, grad-norm 6.547, time 7.18s, 
epoch 220, batches 17000|18183, train-acc 0.865, loss 0.355, para-norm 5099.089, grad-norm 8.094, time 7.10s, 
epoch 220, batches 18000|18183, train-acc 0.869, loss 0.350, para-norm 5098.759, grad-norm 10.026, time 7.16s, 
epoch 220, batches 18183|18183, train-acc 0.859, loss 0.369, para-norm 5098.615, grad-norm 7.909, time 1.30s, 
dev-acc 0.863
epoch 221, batches 1000|18183, train-acc 0.869, loss 0.351, para-norm 5099.981, grad-norm 3.055, time 7.15s, 
epoch 221, batches 2000|18183, train-acc 0.868, loss 0.353, para-norm 5100.593, grad-norm 6.963, time 7.10s, 
epoch 221, batches 3000|18183, train-acc 0.868, loss 0.348, para-norm 5100.635, grad-norm 8.687, time 7.10s, 
epoch 221, batches 4000|18183, train-acc 0.867, loss 0.354, para-norm 5100.192, grad-norm 4.865, time 7.12s, 
epoch 221, batches 5000|18183, train-acc 0.871, loss 0.341, para-norm 5099.962, grad-norm 2.095, time 7.10s, 
epoch 221, batches 6000|18183, train-acc 0.870, loss 0.344, para-norm 5100.504, grad-norm 3.599, time 7.10s, 
epoch 221, batches 7000|18183, train-acc 0.867, loss 0.355, para-norm 5100.419, grad-norm 9.317, time 7.10s, 
epoch 221, batches 8000|18183, train-acc 0.866, loss 0.351, para-norm 5100.316, grad-norm 9.044, time 7.11s, 
epoch 221, batches 9000|18183, train-acc 0.865, loss 0.359, para-norm 5099.705, grad-norm 6.329, time 7.20s, 
epoch 221, batches 10000|18183, train-acc 0.863, loss 0.355, para-norm 5099.111, grad-norm 4.529, time 7.11s, 
epoch 221, batches 11000|18183, train-acc 0.864, loss 0.356, para-norm 5098.913, grad-norm 6.852, time 7.12s, 
epoch 221, batches 12000|18183, train-acc 0.869, loss 0.348, para-norm 5098.440, grad-norm 12.725, time 7.11s, 
epoch 221, batches 13000|18183, train-acc 0.864, loss 0.354, para-norm 5097.953, grad-norm 15.502, time 7.12s, 
epoch 221, batches 14000|18183, train-acc 0.863, loss 0.360, para-norm 5098.111, grad-norm 3.301, time 7.20s, 
epoch 221, batches 15000|18183, train-acc 0.865, loss 0.355, para-norm 5097.620, grad-norm 7.236, time 7.11s, 
epoch 221, batches 16000|18183, train-acc 0.865, loss 0.353, para-norm 5097.596, grad-norm 8.348, time 7.16s, 
epoch 221, batches 17000|18183, train-acc 0.865, loss 0.356, para-norm 5096.909, grad-norm 5.751, time 7.13s, 
epoch 221, batches 18000|18183, train-acc 0.866, loss 0.355, para-norm 5096.821, grad-norm 7.738, time 7.12s, 
epoch 221, batches 18183|18183, train-acc 0.870, loss 0.342, para-norm 5096.802, grad-norm 7.200, time 1.31s, 
dev-acc 0.861
epoch 222, batches 1000|18183, train-acc 0.870, loss 0.347, para-norm 5096.739, grad-norm 6.960, time 7.18s, 
epoch 222, batches 2000|18183, train-acc 0.867, loss 0.351, para-norm 5097.349, grad-norm 6.402, time 7.12s, 
epoch 222, batches 3000|18183, train-acc 0.864, loss 0.355, para-norm 5097.172, grad-norm 17.946, time 7.11s, 
epoch 222, batches 4000|18183, train-acc 0.865, loss 0.350, para-norm 5096.967, grad-norm 13.723, time 7.10s, 
epoch 222, batches 5000|18183, train-acc 0.866, loss 0.355, para-norm 5096.982, grad-norm 5.992, time 7.12s, 
epoch 222, batches 6000|18183, train-acc 0.869, loss 0.344, para-norm 5097.009, grad-norm 5.015, time 7.11s, 
epoch 222, batches 7000|18183, train-acc 0.868, loss 0.349, para-norm 5096.448, grad-norm 331.431, time 7.10s, 
epoch 222, batches 8000|18183, train-acc 0.865, loss 0.352, para-norm 5096.989, grad-norm 11.670, time 7.11s, 
epoch 222, batches 9000|18183, train-acc 0.868, loss 0.349, para-norm 5097.165, grad-norm 5.605, time 7.10s, 
epoch 222, batches 10000|18183, train-acc 0.868, loss 0.347, para-norm 5097.313, grad-norm 5.502, time 7.12s, 
epoch 222, batches 11000|18183, train-acc 0.864, loss 0.356, para-norm 5097.052, grad-norm 3622.741, time 7.12s, 
epoch 222, batches 12000|18183, train-acc 0.862, loss 0.362, para-norm 5096.664, grad-norm 5.251, time 7.10s, 
epoch 222, batches 13000|18183, train-acc 0.867, loss 0.356, para-norm 5095.993, grad-norm 6.814, time 7.10s, 
epoch 222, batches 14000|18183, train-acc 0.865, loss 0.354, para-norm 5095.560, grad-norm 2.003, time 7.10s, 
epoch 222, batches 15000|18183, train-acc 0.863, loss 0.358, para-norm 5094.985, grad-norm 3.352, time 7.11s, 
epoch 222, batches 16000|18183, train-acc 0.868, loss 0.346, para-norm 5095.089, grad-norm 4.104, time 7.09s, 
epoch 222, batches 17000|18183, train-acc 0.863, loss 0.362, para-norm 5094.430, grad-norm 4.029, time 7.11s, 
epoch 222, batches 18000|18183, train-acc 0.863, loss 0.364, para-norm 5094.425, grad-norm 7.998, time 7.09s, 
epoch 222, batches 18183|18183, train-acc 0.860, loss 0.367, para-norm 5094.697, grad-norm 7.359, time 1.30s, 
dev-acc 0.863
epoch 223, batches 1000|18183, train-acc 0.866, loss 0.353, para-norm 5094.916, grad-norm 14.453, time 7.23s, 
epoch 223, batches 2000|18183, train-acc 0.866, loss 0.348, para-norm 5094.188, grad-norm 4.263, time 7.10s, 
epoch 223, batches 3000|18183, train-acc 0.869, loss 0.351, para-norm 5094.053, grad-norm 4.939, time 7.09s, 
epoch 223, batches 4000|18183, train-acc 0.866, loss 0.356, para-norm 5093.815, grad-norm 10.792, time 7.10s, 
epoch 223, batches 5000|18183, train-acc 0.872, loss 0.341, para-norm 5094.422, grad-norm 2.740, time 7.10s, 
epoch 223, batches 6000|18183, train-acc 0.865, loss 0.355, para-norm 5094.489, grad-norm 10.083, time 7.09s, 
epoch 223, batches 7000|18183, train-acc 0.864, loss 0.355, para-norm 5093.535, grad-norm 6.628, time 7.11s, 
epoch 223, batches 8000|18183, train-acc 0.868, loss 0.349, para-norm 5094.155, grad-norm 5.389, time 7.11s, 
epoch 223, batches 9000|18183, train-acc 0.866, loss 0.356, para-norm 5094.267, grad-norm 5.165, time 7.09s, 
epoch 223, batches 10000|18183, train-acc 0.865, loss 0.358, para-norm 5094.245, grad-norm 9.667, time 7.10s, 
epoch 223, batches 11000|18183, train-acc 0.865, loss 0.355, para-norm 5093.996, grad-norm 14.250, time 7.11s, 
epoch 223, batches 12000|18183, train-acc 0.866, loss 0.350, para-norm 5093.637, grad-norm 3.160, time 7.17s, 
epoch 223, batches 13000|18183, train-acc 0.864, loss 0.360, para-norm 5093.151, grad-norm 7.958, time 7.10s, 
epoch 223, batches 14000|18183, train-acc 0.869, loss 0.343, para-norm 5092.812, grad-norm 6.755, time 7.09s, 
epoch 223, batches 15000|18183, train-acc 0.865, loss 0.355, para-norm 5092.257, grad-norm 6.586, time 7.12s, 
epoch 223, batches 16000|18183, train-acc 0.868, loss 0.347, para-norm 5092.062, grad-norm 1.275, time 7.09s, 
epoch 223, batches 17000|18183, train-acc 0.865, loss 0.354, para-norm 5091.928, grad-norm 18.189, time 7.10s, 
epoch 223, batches 18000|18183, train-acc 0.864, loss 0.356, para-norm 5091.726, grad-norm 3.598, time 7.09s, 
epoch 223, batches 18183|18183, train-acc 0.860, loss 0.362, para-norm 5091.825, grad-norm 8.897, time 1.31s, 
dev-acc 0.863
epoch 224, batches 1000|18183, train-acc 0.870, loss 0.346, para-norm 5091.669, grad-norm 4.657, time 7.14s, 
epoch 224, batches 2000|18183, train-acc 0.864, loss 0.354, para-norm 5091.568, grad-norm 9.911, time 7.11s, 
epoch 224, batches 3000|18183, train-acc 0.870, loss 0.345, para-norm 5091.873, grad-norm 0.136, time 7.09s, 
epoch 224, batches 4000|18183, train-acc 0.866, loss 0.352, para-norm 5091.947, grad-norm 10.565, time 7.10s, 
epoch 224, batches 5000|18183, train-acc 0.865, loss 0.355, para-norm 5094.929, grad-norm 11.779, time 7.22s, 
epoch 224, batches 6000|18183, train-acc 0.869, loss 0.346, para-norm 5095.686, grad-norm 6.662, time 7.11s, 
epoch 224, batches 7000|18183, train-acc 0.869, loss 0.346, para-norm 5095.572, grad-norm 13.787, time 7.11s, 
epoch 224, batches 8000|18183, train-acc 0.866, loss 0.356, para-norm 5096.179, grad-norm 56.156, time 7.11s, 
epoch 224, batches 9000|18183, train-acc 0.867, loss 0.353, para-norm 5095.953, grad-norm 5.830, time 7.11s, 
epoch 224, batches 10000|18183, train-acc 0.865, loss 0.352, para-norm 5095.643, grad-norm 2.249, time 7.17s, 
epoch 224, batches 11000|18183, train-acc 0.862, loss 0.361, para-norm 5095.190, grad-norm 2521.513, time 7.13s, 
epoch 224, batches 12000|18183, train-acc 0.866, loss 0.352, para-norm 5094.906, grad-norm 3.711, time 7.11s, 
epoch 224, batches 13000|18183, train-acc 0.863, loss 0.359, para-norm 5094.924, grad-norm 6.832, time 7.09s, 
epoch 224, batches 14000|18183, train-acc 0.867, loss 0.350, para-norm 5094.618, grad-norm 6.515, time 7.11s, 
epoch 224, batches 15000|18183, train-acc 0.868, loss 0.352, para-norm 5095.909, grad-norm 4.760, time 7.10s, 
epoch 224, batches 16000|18183, train-acc 0.868, loss 0.352, para-norm 5095.044, grad-norm 6.309, time 7.10s, 
epoch 224, batches 17000|18183, train-acc 0.864, loss 0.362, para-norm 5094.192, grad-norm 10.638, time 7.13s, 
epoch 224, batches 18000|18183, train-acc 0.863, loss 0.358, para-norm 5094.133, grad-norm 3.317, time 7.12s, 
epoch 224, batches 18183|18183, train-acc 0.873, loss 0.337, para-norm 5094.206, grad-norm 9.184, time 1.32s, 
dev-acc 0.864
epoch 225, batches 1000|18183, train-acc 0.871, loss 0.344, para-norm 5094.426, grad-norm 2.566, time 7.31s, 
epoch 225, batches 2000|18183, train-acc 0.867, loss 0.346, para-norm 5094.355, grad-norm 3.631, time 7.12s, 
epoch 225, batches 3000|18183, train-acc 0.865, loss 0.352, para-norm 5093.835, grad-norm 3.680, time 7.13s, 
epoch 225, batches 4000|18183, train-acc 0.867, loss 0.355, para-norm 5093.774, grad-norm 17.197, time 7.11s, 
epoch 225, batches 5000|18183, train-acc 0.866, loss 0.352, para-norm 5093.409, grad-norm 4.546, time 7.12s, 
epoch 225, batches 6000|18183, train-acc 0.865, loss 0.349, para-norm 5093.910, grad-norm 2.195, time 7.20s, 
epoch 225, batches 7000|18183, train-acc 0.871, loss 0.342, para-norm 5094.053, grad-norm 13.351, time 7.13s, 
epoch 225, batches 8000|18183, train-acc 0.866, loss 0.355, para-norm 5094.007, grad-norm 3.062, time 7.13s, 
epoch 225, batches 9000|18183, train-acc 0.865, loss 0.357, para-norm 5093.381, grad-norm 7.966, time 7.13s, 
epoch 225, batches 10000|18183, train-acc 0.866, loss 0.355, para-norm 5093.243, grad-norm 4.912, time 7.13s, 
epoch 225, batches 11000|18183, train-acc 0.869, loss 0.345, para-norm 5093.023, grad-norm 8.258, time 7.12s, 
epoch 225, batches 12000|18183, train-acc 0.865, loss 0.356, para-norm 5092.908, grad-norm 4.647, time 7.13s, 
epoch 225, batches 13000|18183, train-acc 0.865, loss 0.359, para-norm 5092.282, grad-norm 3.870, time 7.11s, 
epoch 225, batches 14000|18183, train-acc 0.864, loss 0.356, para-norm 5092.452, grad-norm 4.959, time 7.13s, 
epoch 225, batches 15000|18183, train-acc 0.864, loss 0.358, para-norm 5091.715, grad-norm 10.767, time 7.13s, 
epoch 225, batches 16000|18183, train-acc 0.868, loss 0.349, para-norm 5092.156, grad-norm 34.145, time 7.11s, 
epoch 225, batches 17000|18183, train-acc 0.864, loss 0.360, para-norm 5091.895, grad-norm 6.184, time 7.19s, 
epoch 225, batches 18000|18183, train-acc 0.869, loss 0.345, para-norm 5092.057, grad-norm 5.279, time 7.10s, 
epoch 225, batches 18183|18183, train-acc 0.868, loss 0.349, para-norm 5091.862, grad-norm 13.150, time 1.30s, 
dev-acc 0.864
epoch 226, batches 1000|18183, train-acc 0.864, loss 0.352, para-norm 5091.884, grad-norm 4.664, time 7.23s, 
epoch 226, batches 2000|18183, train-acc 0.864, loss 0.359, para-norm 5091.805, grad-norm 9.879, time 7.11s, 
epoch 226, batches 3000|18183, train-acc 0.868, loss 0.349, para-norm 5098.332, grad-norm 7.765, time 7.11s, 
epoch 226, batches 4000|18183, train-acc 0.865, loss 0.354, para-norm 5097.692, grad-norm 15.257, time 7.12s, 
epoch 226, batches 5000|18183, train-acc 0.867, loss 0.356, para-norm 5097.305, grad-norm 7.695, time 7.13s, 
epoch 226, batches 6000|18183, train-acc 0.868, loss 0.350, para-norm 5096.669, grad-norm 9.075, time 7.11s, 
epoch 226, batches 7000|18183, train-acc 0.866, loss 0.350, para-norm 5096.712, grad-norm 8.399, time 7.12s, 
epoch 226, batches 8000|18183, train-acc 0.864, loss 0.356, para-norm 5095.998, grad-norm 8.553, time 7.12s, 
epoch 226, batches 9000|18183, train-acc 0.867, loss 0.348, para-norm 5095.778, grad-norm 9.418, time 7.12s, 
epoch 226, batches 10000|18183, train-acc 0.868, loss 0.351, para-norm 5095.881, grad-norm 4.549, time 7.12s, 
epoch 226, batches 11000|18183, train-acc 0.868, loss 0.350, para-norm 5094.846, grad-norm 8.686, time 7.11s, 
epoch 226, batches 12000|18183, train-acc 0.867, loss 0.353, para-norm 5094.708, grad-norm 3.655, time 7.18s, 
epoch 226, batches 13000|18183, train-acc 0.869, loss 0.350, para-norm 5095.004, grad-norm 4.633, time 7.12s, 
epoch 226, batches 14000|18183, train-acc 0.866, loss 0.354, para-norm 5094.300, grad-norm 6.646, time 7.12s, 
epoch 226, batches 15000|18183, train-acc 0.863, loss 0.352, para-norm 5094.378, grad-norm 11.195, time 7.11s, 
epoch 226, batches 16000|18183, train-acc 0.865, loss 0.353, para-norm 5093.892, grad-norm 6.551, time 7.12s, 
epoch 226, batches 17000|18183, train-acc 0.867, loss 0.350, para-norm 5093.593, grad-norm 7.601, time 7.64s, 
epoch 226, batches 18000|18183, train-acc 0.867, loss 0.345, para-norm 5093.230, grad-norm 5.778, time 8.21s, 
epoch 226, batches 18183|18183, train-acc 0.846, loss 0.383, para-norm 5093.426, grad-norm 9.230, time 1.51s, 
dev-acc 0.863
epoch 227, batches 1000|18183, train-acc 0.868, loss 0.350, para-norm 5093.311, grad-norm 2.657, time 8.48s, 
epoch 227, batches 2000|18183, train-acc 0.866, loss 0.355, para-norm 5093.419, grad-norm 16.272, time 8.21s, 
epoch 227, batches 3000|18183, train-acc 0.868, loss 0.348, para-norm 5093.253, grad-norm 3.612, time 8.22s, 
epoch 227, batches 4000|18183, train-acc 0.870, loss 0.341, para-norm 5092.882, grad-norm 6.144, time 8.27s, 
epoch 227, batches 5000|18183, train-acc 0.868, loss 0.348, para-norm 5092.677, grad-norm 1.611, time 8.32s, 
epoch 227, batches 6000|18183, train-acc 0.867, loss 0.352, para-norm 5092.710, grad-norm 6.339, time 8.37s, 
epoch 227, batches 7000|18183, train-acc 0.868, loss 0.346, para-norm 5092.295, grad-norm 11.349, time 8.21s, 
epoch 227, batches 8000|18183, train-acc 0.869, loss 0.346, para-norm 5092.292, grad-norm 4.502, time 8.22s, 
epoch 227, batches 9000|18183, train-acc 0.864, loss 0.356, para-norm 5092.282, grad-norm 4.242, time 8.20s, 
epoch 227, batches 10000|18183, train-acc 0.868, loss 0.347, para-norm 5098.056, grad-norm 8.413, time 8.20s, 
epoch 227, batches 11000|18183, train-acc 0.864, loss 0.356, para-norm 5096.685, grad-norm 5.771, time 8.19s, 
epoch 227, batches 12000|18183, train-acc 0.865, loss 0.356, para-norm 5096.075, grad-norm 2.381, time 8.19s, 
epoch 227, batches 13000|18183, train-acc 0.870, loss 0.348, para-norm 5095.541, grad-norm 3.131, time 8.18s, 
epoch 227, batches 14000|18183, train-acc 0.865, loss 0.353, para-norm 5095.175, grad-norm 8.047, time 8.19s, 
epoch 227, batches 15000|18183, train-acc 0.864, loss 0.354, para-norm 5094.851, grad-norm 8.744, time 8.20s, 
epoch 227, batches 16000|18183, train-acc 0.867, loss 0.351, para-norm 5095.131, grad-norm 6.038, time 8.19s, 
epoch 227, batches 17000|18183, train-acc 0.866, loss 0.357, para-norm 5095.337, grad-norm 4.545, time 8.18s, 
epoch 227, batches 18000|18183, train-acc 0.863, loss 0.358, para-norm 5094.526, grad-norm 1.642, time 8.21s, 
epoch 227, batches 18183|18183, train-acc 0.864, loss 0.355, para-norm 5094.509, grad-norm 5.825, time 1.50s, 
dev-acc 0.863
epoch 228, batches 1000|18183, train-acc 0.870, loss 0.343, para-norm 5094.867, grad-norm 3.720, time 8.70s, 
epoch 228, batches 2000|18183, train-acc 0.864, loss 0.357, para-norm 5094.586, grad-norm 3.003, time 8.19s, 
epoch 228, batches 3000|18183, train-acc 0.867, loss 0.353, para-norm 5093.954, grad-norm 2.370, time 8.20s, 
epoch 228, batches 4000|18183, train-acc 0.868, loss 0.350, para-norm 5093.756, grad-norm 4.957, time 8.19s, 
epoch 228, batches 5000|18183, train-acc 0.869, loss 0.348, para-norm 5093.716, grad-norm 5.353, time 8.21s, 
epoch 228, batches 6000|18183, train-acc 0.872, loss 0.344, para-norm 5093.513, grad-norm 0.504, time 8.18s, 
epoch 228, batches 7000|18183, train-acc 0.867, loss 0.352, para-norm 5092.907, grad-norm 2.118, time 8.20s, 
epoch 228, batches 8000|18183, train-acc 0.865, loss 0.359, para-norm 5092.873, grad-norm 3.768, time 8.29s, 
epoch 228, batches 9000|18183, train-acc 0.870, loss 0.345, para-norm 5092.572, grad-norm 7.374, time 8.25s, 
epoch 228, batches 10000|18183, train-acc 0.869, loss 0.348, para-norm 5092.780, grad-norm 3.960, time 8.03s, 
epoch 228, batches 11000|18183, train-acc 0.865, loss 0.353, para-norm 5092.058, grad-norm 19.381, time 8.59s, 
epoch 228, batches 12000|18183, train-acc 0.866, loss 0.353, para-norm 5092.300, grad-norm 4.396, time 8.17s, 
epoch 228, batches 13000|18183, train-acc 0.866, loss 0.350, para-norm 5091.901, grad-norm 7.059, time 8.18s, 
epoch 228, batches 14000|18183, train-acc 0.869, loss 0.349, para-norm 5091.068, grad-norm 6.129, time 8.18s, 
epoch 228, batches 15000|18183, train-acc 0.863, loss 0.358, para-norm 5090.822, grad-norm 2.704, time 8.20s, 
epoch 228, batches 16000|18183, train-acc 0.865, loss 0.355, para-norm 5090.531, grad-norm 4.126, time 8.18s, 
epoch 228, batches 17000|18183, train-acc 0.864, loss 0.359, para-norm 5090.671, grad-norm 4.623, time 8.19s, 
epoch 228, batches 18000|18183, train-acc 0.864, loss 0.356, para-norm 5090.677, grad-norm 4.038, time 8.19s, 
epoch 228, batches 18183|18183, train-acc 0.863, loss 0.350, para-norm 5090.503, grad-norm 2.673, time 1.50s, 
dev-acc 0.863
epoch 229, batches 1000|18183, train-acc 0.866, loss 0.353, para-norm 5090.568, grad-norm 191.466, time 8.55s, 
epoch 229, batches 2000|18183, train-acc 0.871, loss 0.342, para-norm 5090.356, grad-norm 9.075, time 8.16s, 
epoch 229, batches 3000|18183, train-acc 0.867, loss 0.349, para-norm 5089.829, grad-norm 9.782, time 8.16s, 
epoch 229, batches 4000|18183, train-acc 0.869, loss 0.346, para-norm 5089.603, grad-norm 9.049, time 8.19s, 
epoch 229, batches 5000|18183, train-acc 0.866, loss 0.353, para-norm 5089.245, grad-norm 19.600, time 8.18s, 
epoch 229, batches 6000|18183, train-acc 0.873, loss 0.340, para-norm 5089.054, grad-norm 7.224, time 8.17s, 
epoch 229, batches 7000|18183, train-acc 0.868, loss 0.349, para-norm 5088.241, grad-norm 5.322, time 8.18s, 
epoch 229, batches 8000|18183, train-acc 0.868, loss 0.347, para-norm 5087.921, grad-norm 7.031, time 8.18s, 
epoch 229, batches 9000|18183, train-acc 0.868, loss 0.348, para-norm 5087.495, grad-norm 11.054, time 8.18s, 
epoch 229, batches 10000|18183, train-acc 0.867, loss 0.352, para-norm 5087.705, grad-norm 2.689, time 8.18s, 
epoch 229, batches 11000|18183, train-acc 0.869, loss 0.347, para-norm 5088.088, grad-norm 3.663, time 8.19s, 
epoch 229, batches 12000|18183, train-acc 0.867, loss 0.352, para-norm 5087.639, grad-norm 4.592, time 8.17s, 
epoch 229, batches 13000|18183, train-acc 0.866, loss 0.358, para-norm 5087.509, grad-norm 0.844, time 8.16s, 
epoch 229, batches 14000|18183, train-acc 0.864, loss 0.357, para-norm 5087.393, grad-norm 5.662, time 8.16s, 
epoch 229, batches 15000|18183, train-acc 0.868, loss 0.351, para-norm 5087.555, grad-norm 6.592, time 8.15s, 
epoch 229, batches 16000|18183, train-acc 0.863, loss 0.360, para-norm 5087.409, grad-norm 6.723, time 8.18s, 
epoch 229, batches 17000|18183, train-acc 0.867, loss 0.355, para-norm 5087.593, grad-norm 1845.202, time 8.18s, 
epoch 229, batches 18000|18183, train-acc 0.865, loss 0.355, para-norm 5087.575, grad-norm 0.845, time 8.17s, 
epoch 229, batches 18183|18183, train-acc 0.864, loss 0.355, para-norm 5087.573, grad-norm 6.882, time 1.50s, 
dev-acc 0.863
epoch 230, batches 1000|18183, train-acc 0.867, loss 0.348, para-norm 5088.164, grad-norm 6.236, time 8.43s, 
epoch 230, batches 2000|18183, train-acc 0.868, loss 0.352, para-norm 5087.676, grad-norm 0.722, time 8.18s, 
epoch 230, batches 3000|18183, train-acc 0.866, loss 0.356, para-norm 5087.155, grad-norm 8.292, time 8.17s, 
epoch 230, batches 4000|18183, train-acc 0.869, loss 0.352, para-norm 5086.392, grad-norm 4.809, time 8.17s, 
epoch 230, batches 5000|18183, train-acc 0.868, loss 0.345, para-norm 5086.634, grad-norm 5.915, time 8.18s, 
epoch 230, batches 6000|18183, train-acc 0.868, loss 0.350, para-norm 5085.523, grad-norm 7.345, time 8.20s, 
epoch 230, batches 7000|18183, train-acc 0.867, loss 0.350, para-norm 5085.681, grad-norm 6.564, time 8.17s, 
epoch 230, batches 8000|18183, train-acc 0.866, loss 0.354, para-norm 5085.987, grad-norm 3.588, time 8.18s, 
epoch 230, batches 9000|18183, train-acc 0.868, loss 0.351, para-norm 5085.250, grad-norm 6.929, time 8.19s, 
epoch 230, batches 10000|18183, train-acc 0.867, loss 0.352, para-norm 5084.757, grad-norm 5.255, time 8.19s, 
epoch 230, batches 11000|18183, train-acc 0.869, loss 0.342, para-norm 5084.799, grad-norm 2.413, time 8.19s, 
epoch 230, batches 12000|18183, train-acc 0.866, loss 0.350, para-norm 5085.447, grad-norm 4.733, time 8.19s, 
epoch 230, batches 13000|18183, train-acc 0.867, loss 0.354, para-norm 5084.512, grad-norm 5.492, time 8.51s, 
epoch 230, batches 14000|18183, train-acc 0.861, loss 0.361, para-norm 5084.446, grad-norm 3.247, time 8.14s, 
epoch 230, batches 15000|18183, train-acc 0.867, loss 0.352, para-norm 5084.165, grad-norm 3.259, time 8.13s, 
epoch 230, batches 16000|18183, train-acc 0.865, loss 0.354, para-norm 5084.369, grad-norm 14.851, time 8.12s, 
epoch 230, batches 17000|18183, train-acc 0.870, loss 0.342, para-norm 5084.508, grad-norm 5.967, time 8.11s, 
epoch 230, batches 18000|18183, train-acc 0.867, loss 0.353, para-norm 5084.790, grad-norm 8.021, time 8.14s, 
epoch 230, batches 18183|18183, train-acc 0.867, loss 0.349, para-norm 5084.464, grad-norm 2.343, time 1.49s, 
dev-acc 0.864
epoch 231, batches 1000|18183, train-acc 0.869, loss 0.349, para-norm 5084.534, grad-norm 4.563, time 8.35s, 
epoch 231, batches 2000|18183, train-acc 0.867, loss 0.352, para-norm 5084.716, grad-norm 21.319, time 8.09s, 
epoch 231, batches 3000|18183, train-acc 0.868, loss 0.350, para-norm 5084.800, grad-norm 5.028, time 8.13s, 
epoch 231, batches 4000|18183, train-acc 0.868, loss 0.349, para-norm 5091.177, grad-norm 5.235, time 8.12s, 
epoch 231, batches 5000|18183, train-acc 0.867, loss 0.351, para-norm 5091.287, grad-norm 24.284, time 8.13s, 
epoch 231, batches 6000|18183, train-acc 0.868, loss 0.350, para-norm 5091.289, grad-norm 6.361, time 8.11s, 
epoch 231, batches 7000|18183, train-acc 0.863, loss 0.356, para-norm 5090.222, grad-norm 6.003, time 8.13s, 
epoch 231, batches 8000|18183, train-acc 0.868, loss 0.353, para-norm 5089.972, grad-norm 7.214, time 8.09s, 
epoch 231, batches 9000|18183, train-acc 0.867, loss 0.353, para-norm 5089.635, grad-norm 0.908, time 8.13s, 
epoch 231, batches 10000|18183, train-acc 0.870, loss 0.347, para-norm 5089.264, grad-norm 5.292, time 8.12s, 
epoch 231, batches 11000|18183, train-acc 0.862, loss 0.359, para-norm 5089.073, grad-norm 4.978, time 8.12s, 
epoch 231, batches 12000|18183, train-acc 0.868, loss 0.347, para-norm 5089.652, grad-norm 403.377, time 8.11s, 
epoch 231, batches 13000|18183, train-acc 0.868, loss 0.350, para-norm 5089.364, grad-norm 3.046, time 8.13s, 
epoch 231, batches 14000|18183, train-acc 0.864, loss 0.355, para-norm 5089.257, grad-norm 4.003, time 8.28s, 
epoch 231, batches 15000|18183, train-acc 0.866, loss 0.351, para-norm 5088.810, grad-norm 7.325, time 8.63s, 
epoch 231, batches 16000|18183, train-acc 0.865, loss 0.352, para-norm 5088.372, grad-norm 3.985, time 7.32s, 
epoch 231, batches 17000|18183, train-acc 0.868, loss 0.346, para-norm 5088.142, grad-norm 5.943, time 7.30s, 
epoch 231, batches 18000|18183, train-acc 0.867, loss 0.351, para-norm 5087.859, grad-norm 15.128, time 7.32s, 
epoch 231, batches 18183|18183, train-acc 0.866, loss 0.356, para-norm 5087.898, grad-norm 9.139, time 1.34s, 
dev-acc 0.863
epoch 232, batches 1000|18183, train-acc 0.872, loss 0.340, para-norm 5087.947, grad-norm 8.201, time 7.36s, 
epoch 232, batches 2000|18183, train-acc 0.873, loss 0.340, para-norm 5087.807, grad-norm 12.301, time 7.32s, 
epoch 232, batches 3000|18183, train-acc 0.866, loss 0.355, para-norm 5087.735, grad-norm 6.475, time 7.31s, 
epoch 232, batches 4000|18183, train-acc 0.870, loss 0.347, para-norm 5087.204, grad-norm 0.000, time 7.31s, 
epoch 232, batches 5000|18183, train-acc 0.869, loss 0.349, para-norm 5087.448, grad-norm 0.604, time 7.31s, 
epoch 232, batches 6000|18183, train-acc 0.866, loss 0.354, para-norm 5087.592, grad-norm 8.774, time 7.30s, 
epoch 232, batches 7000|18183, train-acc 0.863, loss 0.363, para-norm 5086.949, grad-norm 7.489, time 7.31s, 
epoch 232, batches 8000|18183, train-acc 0.866, loss 0.352, para-norm 5086.943, grad-norm 7.499, time 7.30s, 
epoch 232, batches 9000|18183, train-acc 0.868, loss 0.345, para-norm 5086.590, grad-norm 10.141, time 7.30s, 
epoch 232, batches 10000|18183, train-acc 0.866, loss 0.353, para-norm 5086.504, grad-norm 2.949, time 7.32s, 
epoch 232, batches 11000|18183, train-acc 0.866, loss 0.349, para-norm 5086.145, grad-norm 2.751, time 7.31s, 
epoch 232, batches 12000|18183, train-acc 0.865, loss 0.361, para-norm 5085.152, grad-norm 6.803, time 7.32s, 
epoch 232, batches 13000|18183, train-acc 0.870, loss 0.346, para-norm 5085.067, grad-norm 20.703, time 7.31s, 
epoch 232, batches 14000|18183, train-acc 0.865, loss 0.352, para-norm 5084.838, grad-norm 4.510, time 7.32s, 
epoch 232, batches 15000|18183, train-acc 0.867, loss 0.350, para-norm 5084.807, grad-norm 8.055, time 7.32s, 
epoch 232, batches 16000|18183, train-acc 0.862, loss 0.359, para-norm 5083.884, grad-norm 5.376, time 7.33s, 
epoch 232, batches 17000|18183, train-acc 0.868, loss 0.349, para-norm 5083.816, grad-norm 6.582, time 7.32s, 
epoch 232, batches 18000|18183, train-acc 0.864, loss 0.361, para-norm 5083.354, grad-norm 12.517, time 7.32s, 
epoch 232, batches 18183|18183, train-acc 0.861, loss 0.359, para-norm 5083.537, grad-norm 3.343, time 1.34s, 
dev-acc 0.862
epoch 233, batches 1000|18183, train-acc 0.866, loss 0.352, para-norm 5083.715, grad-norm 5.256, time 7.42s, 
epoch 233, batches 2000|18183, train-acc 0.873, loss 0.340, para-norm 5083.536, grad-norm 12.530, time 7.31s, 
epoch 233, batches 3000|18183, train-acc 0.866, loss 0.352, para-norm 5083.309, grad-norm 28.724, time 7.32s, 
epoch 233, batches 4000|18183, train-acc 0.868, loss 0.354, para-norm 5083.181, grad-norm 9.520, time 7.32s, 
epoch 233, batches 5000|18183, train-acc 0.871, loss 0.350, para-norm 5083.298, grad-norm 2.149, time 7.32s, 
epoch 233, batches 6000|18183, train-acc 0.864, loss 0.359, para-norm 5083.463, grad-norm 7.989, time 7.33s, 
epoch 233, batches 7000|18183, train-acc 0.866, loss 0.351, para-norm 5083.750, grad-norm 8.237, time 7.32s, 
epoch 233, batches 8000|18183, train-acc 0.868, loss 0.348, para-norm 5083.607, grad-norm 4.544, time 7.32s, 
epoch 233, batches 9000|18183, train-acc 0.866, loss 0.350, para-norm 5083.227, grad-norm 7.142, time 7.33s, 
epoch 233, batches 10000|18183, train-acc 0.869, loss 0.347, para-norm 5083.278, grad-norm 6.666, time 7.33s, 
epoch 233, batches 11000|18183, train-acc 0.868, loss 0.347, para-norm 5082.938, grad-norm 2.312, time 7.37s, 
epoch 233, batches 12000|18183, train-acc 0.869, loss 0.347, para-norm 5082.219, grad-norm 35.829, time 7.33s, 
epoch 233, batches 13000|18183, train-acc 0.872, loss 0.342, para-norm 5082.543, grad-norm 16.663, time 7.32s, 
epoch 233, batches 14000|18183, train-acc 0.869, loss 0.348, para-norm 5082.787, grad-norm 0.001, time 7.32s, 
epoch 233, batches 15000|18183, train-acc 0.866, loss 0.351, para-norm 5082.614, grad-norm 5.161, time 7.32s, 
epoch 233, batches 16000|18183, train-acc 0.867, loss 0.354, para-norm 5082.277, grad-norm 9.241, time 7.34s, 
epoch 233, batches 17000|18183, train-acc 0.866, loss 0.356, para-norm 5082.444, grad-norm 9.238, time 7.34s, 
epoch 233, batches 18000|18183, train-acc 0.867, loss 0.350, para-norm 5082.994, grad-norm 16.678, time 7.39s, 
epoch 233, batches 18183|18183, train-acc 0.869, loss 0.358, para-norm 5083.082, grad-norm 1.781, time 1.35s, 
dev-acc 0.863
epoch 234, batches 1000|18183, train-acc 0.869, loss 0.347, para-norm 5083.449, grad-norm 10.967, time 7.40s, 
epoch 234, batches 2000|18183, train-acc 0.872, loss 0.340, para-norm 5083.415, grad-norm 2.714, time 7.32s, 
epoch 234, batches 3000|18183, train-acc 0.870, loss 0.347, para-norm 5082.952, grad-norm 3.188, time 7.33s, 
epoch 234, batches 4000|18183, train-acc 0.866, loss 0.355, para-norm 5082.592, grad-norm 6.003, time 7.33s, 
epoch 234, batches 5000|18183, train-acc 0.868, loss 0.352, para-norm 5082.430, grad-norm 15.225, time 7.34s, 
epoch 234, batches 6000|18183, train-acc 0.867, loss 0.357, para-norm 5082.593, grad-norm 2.022, time 7.33s, 
epoch 234, batches 7000|18183, train-acc 0.868, loss 0.351, para-norm 5082.254, grad-norm 3.515, time 7.33s, 
epoch 234, batches 8000|18183, train-acc 0.863, loss 0.359, para-norm 5082.087, grad-norm 4.193, time 7.42s, 
epoch 234, batches 9000|18183, train-acc 0.866, loss 0.350, para-norm 5081.669, grad-norm 15.468, time 7.33s, 
epoch 234, batches 10000|18183, train-acc 0.866, loss 0.351, para-norm 5081.746, grad-norm 4.528, time 7.33s, 
epoch 234, batches 11000|18183, train-acc 0.867, loss 0.354, para-norm 5081.832, grad-norm 3.651, time 7.41s, 
epoch 234, batches 12000|18183, train-acc 0.868, loss 0.351, para-norm 5081.713, grad-norm 5.044, time 7.32s, 
epoch 234, batches 13000|18183, train-acc 0.868, loss 0.349, para-norm 5081.731, grad-norm 5.627, time 7.33s, 
epoch 234, batches 14000|18183, train-acc 0.868, loss 0.348, para-norm 5081.516, grad-norm 5.446, time 7.33s, 
epoch 234, batches 15000|18183, train-acc 0.868, loss 0.350, para-norm 5081.752, grad-norm 11.026, time 7.33s, 
epoch 234, batches 16000|18183, train-acc 0.868, loss 0.348, para-norm 5081.632, grad-norm 2.541, time 7.32s, 
epoch 234, batches 17000|18183, train-acc 0.870, loss 0.343, para-norm 5081.606, grad-norm 8.187, time 7.33s, 
epoch 234, batches 18000|18183, train-acc 0.867, loss 0.351, para-norm 5081.455, grad-norm 14.234, time 7.34s, 
epoch 234, batches 18183|18183, train-acc 0.865, loss 0.365, para-norm 5081.306, grad-norm 7.181, time 1.34s, 
dev-acc 0.862
epoch 235, batches 1000|18183, train-acc 0.866, loss 0.351, para-norm 5081.263, grad-norm 7914.765, time 7.36s, 
epoch 235, batches 2000|18183, train-acc 0.870, loss 0.345, para-norm 5081.434, grad-norm 4.350, time 7.33s, 
epoch 235, batches 3000|18183, train-acc 0.866, loss 0.355, para-norm 5080.490, grad-norm 0.018, time 7.33s, 
epoch 235, batches 4000|18183, train-acc 0.871, loss 0.344, para-norm 5081.062, grad-norm 12.728, time 7.33s, 
epoch 235, batches 5000|18183, train-acc 0.865, loss 0.352, para-norm 5080.465, grad-norm 5.371, time 7.33s, 
epoch 235, batches 6000|18183, train-acc 0.867, loss 0.350, para-norm 5080.745, grad-norm 6.443, time 7.33s, 
epoch 235, batches 7000|18183, train-acc 0.869, loss 0.346, para-norm 5080.537, grad-norm 207.278, time 7.33s, 
epoch 235, batches 8000|18183, train-acc 0.866, loss 0.355, para-norm 5080.040, grad-norm 9.459, time 7.34s, 
epoch 235, batches 9000|18183, train-acc 0.871, loss 0.345, para-norm 5080.269, grad-norm 2.857, time 7.32s, 
epoch 235, batches 10000|18183, train-acc 0.866, loss 0.356, para-norm 5080.264, grad-norm 8.182, time 7.33s, 
epoch 235, batches 11000|18183, train-acc 0.869, loss 0.345, para-norm 5080.561, grad-norm 2.527, time 7.32s, 
epoch 235, batches 12000|18183, train-acc 0.866, loss 0.351, para-norm 5080.558, grad-norm 4.744, time 7.33s, 
epoch 235, batches 13000|18183, train-acc 0.864, loss 0.359, para-norm 5080.205, grad-norm 5.324, time 7.33s, 
epoch 235, batches 14000|18183, train-acc 0.866, loss 0.358, para-norm 5079.742, grad-norm 0.007, time 7.34s, 
epoch 235, batches 15000|18183, train-acc 0.865, loss 0.358, para-norm 5080.182, grad-norm 4.203, time 7.33s, 
epoch 235, batches 16000|18183, train-acc 0.869, loss 0.350, para-norm 5080.205, grad-norm 8.944, time 7.33s, 
epoch 235, batches 17000|18183, train-acc 0.868, loss 0.351, para-norm 5079.634, grad-norm 0.555, time 7.32s, 
epoch 235, batches 18000|18183, train-acc 0.868, loss 0.350, para-norm 5079.284, grad-norm 3.784, time 7.33s, 
epoch 235, batches 18183|18183, train-acc 0.870, loss 0.351, para-norm 5079.344, grad-norm 11.532, time 1.34s, 
dev-acc 0.863
epoch 236, batches 1000|18183, train-acc 0.866, loss 0.355, para-norm 5079.106, grad-norm 3.811, time 7.38s, 
epoch 236, batches 2000|18183, train-acc 0.866, loss 0.353, para-norm 5078.798, grad-norm 6.550, time 7.32s, 
epoch 236, batches 3000|18183, train-acc 0.873, loss 0.341, para-norm 5078.671, grad-norm 4.240, time 7.33s, 
epoch 236, batches 4000|18183, train-acc 0.869, loss 0.345, para-norm 5078.842, grad-norm 5.129, time 7.33s, 
epoch 236, batches 5000|18183, train-acc 0.870, loss 0.341, para-norm 5078.830, grad-norm 2.758, time 7.32s, 
epoch 236, batches 6000|18183, train-acc 0.869, loss 0.347, para-norm 5079.274, grad-norm 5.683, time 7.33s, 
epoch 236, batches 7000|18183, train-acc 0.867, loss 0.350, para-norm 5079.105, grad-norm 3.068, time 7.33s, 
epoch 236, batches 8000|18183, train-acc 0.865, loss 0.357, para-norm 5079.007, grad-norm 9.344, time 7.34s, 
epoch 236, batches 9000|18183, train-acc 0.869, loss 0.342, para-norm 5078.904, grad-norm 8.482, time 7.40s, 
epoch 236, batches 10000|18183, train-acc 0.869, loss 0.348, para-norm 5078.802, grad-norm 2.781, time 7.32s, 
epoch 236, batches 11000|18183, train-acc 0.867, loss 0.349, para-norm 5084.266, grad-norm 3.170, time 7.33s, 
epoch 236, batches 12000|18183, train-acc 0.865, loss 0.355, para-norm 5084.651, grad-norm 5.156, time 7.34s, 
epoch 236, batches 13000|18183, train-acc 0.865, loss 0.354, para-norm 5083.744, grad-norm 9.900, time 7.32s, 
epoch 236, batches 14000|18183, train-acc 0.863, loss 0.356, para-norm 5083.499, grad-norm 5.928, time 7.42s, 
epoch 236, batches 15000|18183, train-acc 0.867, loss 0.350, para-norm 5082.814, grad-norm 2.779, time 7.32s, 
epoch 236, batches 16000|18183, train-acc 0.865, loss 0.350, para-norm 5082.807, grad-norm 8.107, time 7.33s, 
epoch 236, batches 17000|18183, train-acc 0.866, loss 0.355, para-norm 5081.785, grad-norm 5.183, time 7.33s, 
epoch 236, batches 18000|18183, train-acc 0.868, loss 0.348, para-norm 5081.500, grad-norm 11.275, time 7.33s, 
epoch 236, batches 18183|18183, train-acc 0.861, loss 0.368, para-norm 5081.451, grad-norm 6.882, time 1.34s, 
dev-acc 0.862
epoch 237, batches 1000|18183, train-acc 0.867, loss 0.350, para-norm 5081.011, grad-norm 1.861, time 7.40s, 
epoch 237, batches 2000|18183, train-acc 0.873, loss 0.338, para-norm 5081.342, grad-norm 9.850, time 7.33s, 
epoch 237, batches 3000|18183, train-acc 0.865, loss 0.350, para-norm 5080.931, grad-norm 28.131, time 7.35s, 
epoch 237, batches 4000|18183, train-acc 0.871, loss 0.339, para-norm 5081.268, grad-norm 6.016, time 7.32s, 
epoch 237, batches 5000|18183, train-acc 0.870, loss 0.347, para-norm 5080.482, grad-norm 5.265, time 7.33s, 
epoch 237, batches 6000|18183, train-acc 0.867, loss 0.350, para-norm 5080.362, grad-norm 10.703, time 7.34s, 
epoch 237, batches 7000|18183, train-acc 0.871, loss 0.349, para-norm 5080.010, grad-norm 14.824, time 7.32s, 
epoch 237, batches 8000|18183, train-acc 0.866, loss 0.352, para-norm 5079.589, grad-norm 12.816, time 7.33s, 
epoch 237, batches 9000|18183, train-acc 0.867, loss 0.350, para-norm 5079.171, grad-norm 5.532, time 7.32s, 
epoch 237, batches 10000|18183, train-acc 0.866, loss 0.356, para-norm 5078.929, grad-norm 4.418, time 7.33s, 
epoch 237, batches 11000|18183, train-acc 0.867, loss 0.353, para-norm 5078.823, grad-norm 9.878, time 7.33s, 
epoch 237, batches 12000|18183, train-acc 0.866, loss 0.354, para-norm 5078.497, grad-norm 15.326, time 7.32s, 
epoch 237, batches 13000|18183, train-acc 0.867, loss 0.351, para-norm 5077.772, grad-norm 6.574, time 7.34s, 
epoch 237, batches 14000|18183, train-acc 0.864, loss 0.355, para-norm 5078.127, grad-norm 44.870, time 7.32s, 
epoch 237, batches 15000|18183, train-acc 0.870, loss 0.345, para-norm 5077.988, grad-norm 17.199, time 7.33s, 
epoch 237, batches 16000|18183, train-acc 0.867, loss 0.355, para-norm 5077.419, grad-norm 15.127, time 7.34s, 
epoch 237, batches 17000|18183, train-acc 0.868, loss 0.351, para-norm 5077.377, grad-norm 1.299, time 7.33s, 
epoch 237, batches 18000|18183, train-acc 0.866, loss 0.353, para-norm 5077.495, grad-norm 3.884, time 7.33s, 
epoch 237, batches 18183|18183, train-acc 0.874, loss 0.338, para-norm 5077.426, grad-norm 10.785, time 1.35s, 
dev-acc 0.862
epoch 238, batches 1000|18183, train-acc 0.870, loss 0.342, para-norm 5077.156, grad-norm 3.411, time 7.39s, 
epoch 238, batches 2000|18183, train-acc 0.873, loss 0.344, para-norm 5077.017, grad-norm 4.357, time 7.32s, 
epoch 238, batches 3000|18183, train-acc 0.869, loss 0.346, para-norm 5076.994, grad-norm 467.099, time 7.35s, 
epoch 238, batches 4000|18183, train-acc 0.864, loss 0.357, para-norm 5077.274, grad-norm 6.497, time 7.34s, 
epoch 238, batches 5000|18183, train-acc 0.871, loss 0.342, para-norm 5077.397, grad-norm 15.855, time 7.33s, 
epoch 238, batches 6000|18183, train-acc 0.870, loss 0.342, para-norm 5077.220, grad-norm 6.287, time 7.33s, 
epoch 238, batches 7000|18183, train-acc 0.866, loss 0.353, para-norm 5076.821, grad-norm 6.735, time 7.33s, 
epoch 238, batches 8000|18183, train-acc 0.870, loss 0.344, para-norm 5076.746, grad-norm 52.006, time 7.34s, 
epoch 238, batches 9000|18183, train-acc 0.865, loss 0.348, para-norm 5077.134, grad-norm 3.124, time 7.33s, 
epoch 238, batches 10000|18183, train-acc 0.869, loss 0.346, para-norm 5077.137, grad-norm 13.255, time 7.32s, 
epoch 238, batches 11000|18183, train-acc 0.871, loss 0.344, para-norm 5076.946, grad-norm 11.742, time 7.33s, 
epoch 238, batches 12000|18183, train-acc 0.862, loss 0.358, para-norm 5077.189, grad-norm 8.587, time 7.34s, 
epoch 238, batches 13000|18183, train-acc 0.870, loss 0.349, para-norm 5076.227, grad-norm 6.801, time 7.33s, 
epoch 238, batches 14000|18183, train-acc 0.866, loss 0.355, para-norm 5075.547, grad-norm 7.089, time 7.33s, 
epoch 238, batches 15000|18183, train-acc 0.865, loss 0.355, para-norm 5075.483, grad-norm 3.564, time 7.33s, 
epoch 238, batches 16000|18183, train-acc 0.865, loss 0.358, para-norm 5075.527, grad-norm 3.746, time 7.32s, 
epoch 238, batches 17000|18183, train-acc 0.866, loss 0.352, para-norm 5075.609, grad-norm 11.815, time 7.32s, 
epoch 238, batches 18000|18183, train-acc 0.867, loss 0.348, para-norm 5075.692, grad-norm 2.431, time 7.34s, 
epoch 238, batches 18183|18183, train-acc 0.860, loss 0.368, para-norm 5075.754, grad-norm 5.958, time 1.34s, 
dev-acc 0.862
epoch 239, batches 1000|18183, train-acc 0.866, loss 0.355, para-norm 5075.521, grad-norm 15.254, time 7.39s, 
epoch 239, batches 2000|18183, train-acc 0.868, loss 0.345, para-norm 5075.073, grad-norm 4.955, time 7.33s, 
epoch 239, batches 3000|18183, train-acc 0.864, loss 0.354, para-norm 5075.107, grad-norm 1.213, time 7.33s, 
epoch 239, batches 4000|18183, train-acc 0.869, loss 0.348, para-norm 5075.206, grad-norm 0.193, time 7.33s, 
epoch 239, batches 5000|18183, train-acc 0.870, loss 0.344, para-norm 5075.025, grad-norm 8.601, time 7.33s, 
epoch 239, batches 6000|18183, train-acc 0.865, loss 0.354, para-norm 5074.639, grad-norm 2.467, time 7.33s, 
epoch 239, batches 7000|18183, train-acc 0.869, loss 0.345, para-norm 5074.448, grad-norm 7.200, time 7.34s, 
epoch 239, batches 8000|18183, train-acc 0.866, loss 0.348, para-norm 5074.845, grad-norm 2.711, time 7.33s, 
epoch 239, batches 9000|18183, train-acc 0.870, loss 0.347, para-norm 5074.634, grad-norm 7.277, time 7.33s, 
epoch 239, batches 10000|18183, train-acc 0.864, loss 0.358, para-norm 5074.304, grad-norm 7.309, time 7.34s, 
epoch 239, batches 11000|18183, train-acc 0.866, loss 0.349, para-norm 5074.139, grad-norm 6.093, time 7.33s, 
epoch 239, batches 12000|18183, train-acc 0.868, loss 0.348, para-norm 5073.656, grad-norm 9.660, time 7.37s, 
epoch 239, batches 13000|18183, train-acc 0.865, loss 0.357, para-norm 5073.070, grad-norm 8.224, time 7.08s, 
epoch 239, batches 14000|18183, train-acc 0.867, loss 0.351, para-norm 5072.687, grad-norm 10.847, time 7.06s, 
epoch 239, batches 15000|18183, train-acc 0.869, loss 0.346, para-norm 5072.819, grad-norm 40.250, time 7.07s, 
epoch 239, batches 16000|18183, train-acc 0.868, loss 0.351, para-norm 5072.474, grad-norm 5.058, time 7.05s, 
epoch 239, batches 17000|18183, train-acc 0.869, loss 0.344, para-norm 5072.672, grad-norm 7.776, time 7.06s, 
epoch 239, batches 18000|18183, train-acc 0.868, loss 0.351, para-norm 5071.856, grad-norm 4.317, time 7.07s, 
epoch 239, batches 18183|18183, train-acc 0.868, loss 0.353, para-norm 5072.194, grad-norm 3.724, time 1.29s, 
dev-acc 0.863
epoch 240, batches 1000|18183, train-acc 0.870, loss 0.340, para-norm 5072.310, grad-norm 14.873, time 7.22s, 
epoch 240, batches 2000|18183, train-acc 0.867, loss 0.355, para-norm 5072.040, grad-norm 10.317, time 7.05s, 
epoch 240, batches 3000|18183, train-acc 0.871, loss 0.336, para-norm 5071.996, grad-norm 4.612, time 7.06s, 
epoch 240, batches 4000|18183, train-acc 0.870, loss 0.349, para-norm 5073.086, grad-norm 13.505, time 7.07s, 
epoch 240, batches 5000|18183, train-acc 0.872, loss 0.338, para-norm 5073.612, grad-norm 2.300, time 7.07s, 
epoch 240, batches 6000|18183, train-acc 0.865, loss 0.359, para-norm 5073.034, grad-norm 5.944, time 7.09s, 
epoch 240, batches 7000|18183, train-acc 0.867, loss 0.350, para-norm 5073.288, grad-norm 3.418, time 7.08s, 
epoch 240, batches 8000|18183, train-acc 0.865, loss 0.352, para-norm 5072.941, grad-norm 5.823, time 7.05s, 
epoch 240, batches 9000|18183, train-acc 0.870, loss 0.349, para-norm 5072.989, grad-norm 6.317, time 7.04s, 
epoch 240, batches 10000|18183, train-acc 0.866, loss 0.354, para-norm 5072.248, grad-norm 7.417, time 7.06s, 
epoch 240, batches 11000|18183, train-acc 0.871, loss 0.346, para-norm 5071.853, grad-norm 9.812, time 7.05s, 
epoch 240, batches 12000|18183, train-acc 0.868, loss 0.350, para-norm 5071.823, grad-norm 2.556, time 7.08s, 
epoch 240, batches 13000|18183, train-acc 0.866, loss 0.355, para-norm 5071.649, grad-norm 4.413, time 7.07s, 
epoch 240, batches 14000|18183, train-acc 0.863, loss 0.360, para-norm 5071.571, grad-norm 4.028, time 7.08s, 
epoch 240, batches 15000|18183, train-acc 0.867, loss 0.351, para-norm 5071.942, grad-norm 1.111, time 7.06s, 
epoch 240, batches 16000|18183, train-acc 0.868, loss 0.348, para-norm 5071.241, grad-norm 6.838, time 7.05s, 
epoch 240, batches 17000|18183, train-acc 0.864, loss 0.356, para-norm 5071.124, grad-norm 3.117, time 7.05s, 
epoch 240, batches 18000|18183, train-acc 0.867, loss 0.351, para-norm 5071.437, grad-norm 1.823, time 7.05s, 
epoch 240, batches 18183|18183, train-acc 0.863, loss 0.359, para-norm 5071.190, grad-norm 7.248, time 1.29s, 
dev-acc 0.864
epoch 241, batches 1000|18183, train-acc 0.868, loss 0.350, para-norm 5071.510, grad-norm 5.582, time 7.16s, 
epoch 241, batches 2000|18183, train-acc 0.868, loss 0.353, para-norm 5071.534, grad-norm 13.926, time 7.08s, 
epoch 241, batches 3000|18183, train-acc 0.866, loss 0.353, para-norm 5071.574, grad-norm 1.931, time 7.19s, 
epoch 241, batches 4000|18183, train-acc 0.870, loss 0.348, para-norm 5071.558, grad-norm 2.268, time 7.05s, 
epoch 241, batches 5000|18183, train-acc 0.870, loss 0.344, para-norm 5071.545, grad-norm 5.259, time 7.04s, 
epoch 241, batches 6000|18183, train-acc 0.872, loss 0.341, para-norm 5070.945, grad-norm 9.504, time 7.91s, 
epoch 241, batches 7000|18183, train-acc 0.866, loss 0.346, para-norm 5070.949, grad-norm 386.757, time 7.98s, 
epoch 241, batches 8000|18183, train-acc 0.869, loss 0.343, para-norm 5071.202, grad-norm 4.717, time 7.92s, 
epoch 241, batches 9000|18183, train-acc 0.870, loss 0.342, para-norm 5071.471, grad-norm 2.799, time 7.94s, 
epoch 241, batches 10000|18183, train-acc 0.866, loss 0.349, para-norm 5071.328, grad-norm 9.350, time 7.93s, 
epoch 241, batches 11000|18183, train-acc 0.871, loss 0.342, para-norm 5071.500, grad-norm 7.246, time 7.93s, 
epoch 241, batches 12000|18183, train-acc 0.867, loss 0.351, para-norm 5071.127, grad-norm 7.444, time 7.94s, 
epoch 241, batches 13000|18183, train-acc 0.868, loss 0.346, para-norm 5071.262, grad-norm 5.889, time 7.91s, 
epoch 241, batches 14000|18183, train-acc 0.865, loss 0.355, para-norm 5070.709, grad-norm 4.782, time 7.95s, 
epoch 241, batches 15000|18183, train-acc 0.868, loss 0.351, para-norm 5070.989, grad-norm 2.413, time 7.92s, 
epoch 241, batches 16000|18183, train-acc 0.869, loss 0.350, para-norm 5070.674, grad-norm 2.833, time 7.93s, 
epoch 241, batches 17000|18183, train-acc 0.863, loss 0.362, para-norm 5070.204, grad-norm 14.553, time 7.94s, 
epoch 241, batches 18000|18183, train-acc 0.866, loss 0.355, para-norm 5070.050, grad-norm 10.143, time 7.91s, 
epoch 241, batches 18183|18183, train-acc 0.860, loss 0.377, para-norm 5069.916, grad-norm 2596.346, time 1.46s, 
dev-acc 0.863
epoch 242, batches 1000|18183, train-acc 0.872, loss 0.344, para-norm 5070.427, grad-norm 7.479, time 7.97s, 
epoch 242, batches 2000|18183, train-acc 0.868, loss 0.346, para-norm 5070.807, grad-norm 4.575, time 7.93s, 
epoch 242, batches 3000|18183, train-acc 0.870, loss 0.343, para-norm 5070.571, grad-norm 5.592, time 7.93s, 
epoch 242, batches 4000|18183, train-acc 0.870, loss 0.347, para-norm 5070.278, grad-norm 3.971, time 7.93s, 
epoch 242, batches 5000|18183, train-acc 0.869, loss 0.346, para-norm 5070.686, grad-norm 5.477, time 7.93s, 
epoch 242, batches 6000|18183, train-acc 0.867, loss 0.352, para-norm 5070.272, grad-norm 7.460, time 7.93s, 
epoch 242, batches 7000|18183, train-acc 0.867, loss 0.349, para-norm 5073.965, grad-norm 9.873, time 7.92s, 
epoch 242, batches 8000|18183, train-acc 0.869, loss 0.350, para-norm 5073.944, grad-norm 9.537, time 7.94s, 
epoch 242, batches 9000|18183, train-acc 0.869, loss 0.350, para-norm 5073.815, grad-norm 7.734, time 7.93s, 
epoch 242, batches 10000|18183, train-acc 0.869, loss 0.350, para-norm 5073.355, grad-norm 12.524, time 7.94s, 
epoch 242, batches 11000|18183, train-acc 0.872, loss 0.342, para-norm 5072.930, grad-norm 5.421, time 8.00s, 
epoch 242, batches 12000|18183, train-acc 0.867, loss 0.354, para-norm 5072.400, grad-norm 2.382, time 7.39s, 
epoch 242, batches 13000|18183, train-acc 0.869, loss 0.348, para-norm 5072.382, grad-norm 3.372, time 7.14s, 
epoch 242, batches 14000|18183, train-acc 0.867, loss 0.350, para-norm 5072.362, grad-norm 4.876, time 7.07s, 
epoch 242, batches 15000|18183, train-acc 0.867, loss 0.350, para-norm 5072.294, grad-norm 2.565, time 7.08s, 
epoch 242, batches 16000|18183, train-acc 0.865, loss 0.354, para-norm 5072.086, grad-norm 5.607, time 7.07s, 
epoch 242, batches 17000|18183, train-acc 0.865, loss 0.352, para-norm 5071.603, grad-norm 3.084, time 7.08s, 
epoch 242, batches 18000|18183, train-acc 0.868, loss 0.346, para-norm 5071.726, grad-norm 6.077, time 7.07s, 
epoch 242, batches 18183|18183, train-acc 0.870, loss 0.342, para-norm 5071.592, grad-norm 6.946, time 1.30s, 
dev-acc 0.863
epoch 243, batches 1000|18183, train-acc 0.871, loss 0.346, para-norm 5071.632, grad-norm 6.239, time 7.16s, 
epoch 243, batches 2000|18183, train-acc 0.867, loss 0.351, para-norm 5071.467, grad-norm 13.159, time 7.08s, 
epoch 243, batches 3000|18183, train-acc 0.868, loss 0.346, para-norm 5071.666, grad-norm 5.728, time 7.07s, 
epoch 243, batches 4000|18183, train-acc 0.869, loss 0.345, para-norm 5072.185, grad-norm 7.031, time 7.06s, 
epoch 243, batches 5000|18183, train-acc 0.866, loss 0.354, para-norm 5071.850, grad-norm 8.438, time 7.07s, 
epoch 243, batches 6000|18183, train-acc 0.867, loss 0.350, para-norm 5071.112, grad-norm 5.991, time 7.07s, 
epoch 243, batches 7000|18183, train-acc 0.870, loss 0.346, para-norm 5071.346, grad-norm 8.086, time 7.14s, 
epoch 243, batches 8000|18183, train-acc 0.868, loss 0.349, para-norm 5071.139, grad-norm 4.115, time 7.07s, 
epoch 243, batches 9000|18183, train-acc 0.866, loss 0.351, para-norm 5070.736, grad-norm 5.609, time 7.08s, 
epoch 243, batches 10000|18183, train-acc 0.870, loss 0.343, para-norm 5071.143, grad-norm 5.095, time 7.07s, 
epoch 243, batches 11000|18183, train-acc 0.870, loss 0.345, para-norm 5070.720, grad-norm 9.468, time 7.07s, 
epoch 243, batches 12000|18183, train-acc 0.870, loss 0.347, para-norm 5070.120, grad-norm 12.168, time 7.07s, 
epoch 243, batches 13000|18183, train-acc 0.862, loss 0.360, para-norm 5069.944, grad-norm 5.939, time 7.09s, 
epoch 243, batches 14000|18183, train-acc 0.866, loss 0.349, para-norm 5069.260, grad-norm 4.695, time 7.08s, 
epoch 243, batches 15000|18183, train-acc 0.870, loss 0.341, para-norm 5069.752, grad-norm 2.487, time 7.07s, 
epoch 243, batches 16000|18183, train-acc 0.864, loss 0.359, para-norm 5069.385, grad-norm 4.028, time 7.06s, 
epoch 243, batches 17000|18183, train-acc 0.870, loss 0.346, para-norm 5068.985, grad-norm 7.414, time 7.07s, 
epoch 243, batches 18000|18183, train-acc 0.865, loss 0.355, para-norm 5068.821, grad-norm 4.462, time 7.08s, 
epoch 243, batches 18183|18183, train-acc 0.869, loss 0.353, para-norm 5068.991, grad-norm 5.185, time 1.29s, 
dev-acc 0.862
epoch 244, batches 1000|18183, train-acc 0.869, loss 0.346, para-norm 5069.187, grad-norm 4.233, time 7.12s, 
epoch 244, batches 2000|18183, train-acc 0.869, loss 0.343, para-norm 5069.096, grad-norm 6.018, time 7.08s, 
epoch 244, batches 3000|18183, train-acc 0.867, loss 0.349, para-norm 5068.932, grad-norm 20.905, time 7.08s, 
epoch 244, batches 4000|18183, train-acc 0.869, loss 0.351, para-norm 5068.927, grad-norm 8.899, time 7.07s, 
epoch 244, batches 5000|18183, train-acc 0.868, loss 0.351, para-norm 5068.645, grad-norm 12.801, time 7.16s, 
epoch 244, batches 6000|18183, train-acc 0.869, loss 0.348, para-norm 5068.542, grad-norm 8.830, time 7.08s, 
epoch 244, batches 7000|18183, train-acc 0.872, loss 0.344, para-norm 5068.092, grad-norm 14.283, time 7.09s, 
epoch 244, batches 8000|18183, train-acc 0.869, loss 0.347, para-norm 5068.291, grad-norm 8.817, time 7.07s, 
epoch 244, batches 9000|18183, train-acc 0.867, loss 0.352, para-norm 5068.124, grad-norm 5.434, time 7.08s, 
epoch 244, batches 10000|18183, train-acc 0.868, loss 0.348, para-norm 5067.743, grad-norm 2.081, time 7.08s, 
epoch 244, batches 11000|18183, train-acc 0.866, loss 0.351, para-norm 5067.845, grad-norm 3.629, time 7.08s, 
epoch 244, batches 12000|18183, train-acc 0.864, loss 0.360, para-norm 5067.178, grad-norm 6.571, time 7.07s, 
epoch 244, batches 13000|18183, train-acc 0.868, loss 0.348, para-norm 5066.796, grad-norm 9.812, time 7.07s, 
epoch 244, batches 14000|18183, train-acc 0.867, loss 0.352, para-norm 5067.587, grad-norm 5.447, time 7.07s, 
epoch 244, batches 15000|18183, train-acc 0.868, loss 0.350, para-norm 5067.102, grad-norm 5.586, time 7.08s, 
epoch 244, batches 16000|18183, train-acc 0.869, loss 0.344, para-norm 5067.017, grad-norm 7.075, time 7.06s, 
epoch 244, batches 17000|18183, train-acc 0.869, loss 0.346, para-norm 5067.419, grad-norm 20.763, time 7.08s, 
epoch 244, batches 18000|18183, train-acc 0.865, loss 0.354, para-norm 5066.814, grad-norm 8.000, time 7.08s, 
epoch 244, batches 18183|18183, train-acc 0.868, loss 0.355, para-norm 5067.037, grad-norm 9.257, time 1.30s, 
dev-acc 0.863
epoch 245, batches 1000|18183, train-acc 0.868, loss 0.346, para-norm 5067.259, grad-norm 2.823, time 7.11s, 
epoch 245, batches 2000|18183, train-acc 0.870, loss 0.344, para-norm 5067.800, grad-norm 7.450, time 7.24s, 
epoch 245, batches 3000|18183, train-acc 0.870, loss 0.346, para-norm 5067.688, grad-norm 4.495, time 7.35s, 
epoch 245, batches 4000|18183, train-acc 0.868, loss 0.352, para-norm 5066.887, grad-norm 4.969, time 7.34s, 
epoch 245, batches 5000|18183, train-acc 0.869, loss 0.351, para-norm 5066.524, grad-norm 14.161, time 7.33s, 
epoch 245, batches 6000|18183, train-acc 0.867, loss 0.346, para-norm 5066.015, grad-norm 4.716, time 7.34s, 
epoch 245, batches 7000|18183, train-acc 0.869, loss 0.352, para-norm 5066.141, grad-norm 0.000, time 7.34s, 
epoch 245, batches 8000|18183, train-acc 0.869, loss 0.343, para-norm 5066.266, grad-norm 5.984, time 7.34s, 
epoch 245, batches 9000|18183, train-acc 0.864, loss 0.355, para-norm 5066.313, grad-norm 16.128, time 7.34s, 
epoch 245, batches 10000|18183, train-acc 0.868, loss 0.349, para-norm 5066.450, grad-norm 5.368, time 7.35s, 
epoch 245, batches 11000|18183, train-acc 0.867, loss 0.348, para-norm 5066.027, grad-norm 11.780, time 7.34s, 
epoch 245, batches 12000|18183, train-acc 0.868, loss 0.346, para-norm 5065.663, grad-norm 3.637, time 7.34s, 
epoch 245, batches 13000|18183, train-acc 0.868, loss 0.347, para-norm 5065.895, grad-norm 9.834, time 7.33s, 
epoch 245, batches 14000|18183, train-acc 0.866, loss 0.349, para-norm 5065.978, grad-norm 9.454, time 7.34s, 
epoch 245, batches 15000|18183, train-acc 0.865, loss 0.356, para-norm 5065.289, grad-norm 11.727, time 7.34s, 
epoch 245, batches 16000|18183, train-acc 0.868, loss 0.346, para-norm 5065.334, grad-norm 22.650, time 7.33s, 
epoch 245, batches 17000|18183, train-acc 0.871, loss 0.343, para-norm 5065.325, grad-norm 7.303, time 7.33s, 
epoch 245, batches 18000|18183, train-acc 0.868, loss 0.348, para-norm 5065.246, grad-norm 13.461, time 7.34s, 
epoch 245, batches 18183|18183, train-acc 0.869, loss 0.343, para-norm 5065.232, grad-norm 2.356, time 1.34s, 
dev-acc 0.865
current best-dev:
	0 0.651
	1 0.682
	2 0.700
	3 0.717
	4 0.720
	5 0.731
	6 0.738
	7 0.747
	8 0.755
	9 0.757
	10 0.767
	11 0.770
	12 0.772
	13 0.774
	14 0.776
	15 0.784
	17 0.789
	19 0.790
	20 0.793
	21 0.793
	22 0.794
	24 0.795
	26 0.802
	28 0.806
	33 0.807
	35 0.812
	41 0.814
	43 0.815
	45 0.818
	47 0.821
	52 0.822
	53 0.823
	54 0.826
	58 0.829
	60 0.831
	61 0.833
	63 0.833
	64 0.837
	66 0.842
	73 0.845
	75 0.846
	79 0.847
	80 0.848
	81 0.849
	83 0.850
	88 0.853
	93 0.854
	96 0.857
	109 0.857
	111 0.857
	115 0.857
	120 0.859
	126 0.860
	142 0.860
	147 0.862
	152 0.862
	185 0.863
	203 0.864
	204 0.864
	245 0.865
save model!
epoch 246, batches 1000|18183, train-acc 0.867, loss 0.347, para-norm 5065.410, grad-norm 9.459, time 7.45s, 
epoch 246, batches 2000|18183, train-acc 0.869, loss 0.346, para-norm 5064.904, grad-norm 3.083, time 7.33s, 
epoch 246, batches 3000|18183, train-acc 0.869, loss 0.347, para-norm 5065.044, grad-norm 4.258, time 7.33s, 
epoch 246, batches 4000|18183, train-acc 0.870, loss 0.343, para-norm 5064.739, grad-norm 7.841, time 7.33s, 
epoch 246, batches 5000|18183, train-acc 0.868, loss 0.349, para-norm 5064.887, grad-norm 1.620, time 7.35s, 
epoch 246, batches 6000|18183, train-acc 0.869, loss 0.348, para-norm 5064.658, grad-norm 22.482, time 7.34s, 
epoch 246, batches 7000|18183, train-acc 0.869, loss 0.347, para-norm 5064.579, grad-norm 6.536, time 7.20s, 
epoch 246, batches 8000|18183, train-acc 0.868, loss 0.351, para-norm 5064.657, grad-norm 4.722, time 7.05s, 
epoch 246, batches 9000|18183, train-acc 0.871, loss 0.344, para-norm 5064.190, grad-norm 98.862, time 7.05s, 
epoch 246, batches 10000|18183, train-acc 0.865, loss 0.354, para-norm 5064.451, grad-norm 16.233, time 7.07s, 
epoch 246, batches 11000|18183, train-acc 0.865, loss 0.357, para-norm 5063.801, grad-norm 2.804, time 7.07s, 
epoch 246, batches 12000|18183, train-acc 0.864, loss 0.356, para-norm 5063.119, grad-norm 9.985, time 7.08s, 
epoch 246, batches 13000|18183, train-acc 0.869, loss 0.349, para-norm 5062.939, grad-norm 8.682, time 7.06s, 
epoch 246, batches 14000|18183, train-acc 0.867, loss 0.348, para-norm 5062.300, grad-norm 12.629, time 7.07s, 
epoch 246, batches 15000|18183, train-acc 0.870, loss 0.350, para-norm 5062.187, grad-norm 5.556, time 7.06s, 
epoch 246, batches 16000|18183, train-acc 0.865, loss 0.356, para-norm 5062.369, grad-norm 3.286, time 7.08s, 
epoch 246, batches 17000|18183, train-acc 0.866, loss 0.355, para-norm 5062.231, grad-norm 8.903, time 7.08s, 
epoch 246, batches 18000|18183, train-acc 0.866, loss 0.353, para-norm 5061.772, grad-norm 3.898, time 7.07s, 
epoch 246, batches 18183|18183, train-acc 0.882, loss 0.324, para-norm 5061.861, grad-norm 2.545, time 1.29s, 
dev-acc 0.863
epoch 247, batches 1000|18183, train-acc 0.870, loss 0.342, para-norm 5061.891, grad-norm 1.461, time 7.13s, 
epoch 247, batches 2000|18183, train-acc 0.870, loss 0.345, para-norm 5062.149, grad-norm 4.113, time 7.06s, 
epoch 247, batches 3000|18183, train-acc 0.871, loss 0.339, para-norm 5062.863, grad-norm 6.524, time 7.07s, 
epoch 247, batches 4000|18183, train-acc 0.869, loss 0.349, para-norm 5062.632, grad-norm 4.084, time 7.06s, 
epoch 247, batches 5000|18183, train-acc 0.869, loss 0.344, para-norm 5062.077, grad-norm 7.171, time 7.06s, 
epoch 247, batches 6000|18183, train-acc 0.863, loss 0.358, para-norm 5062.210, grad-norm 3.031, time 7.07s, 
epoch 247, batches 7000|18183, train-acc 0.866, loss 0.351, para-norm 5062.491, grad-norm 1.543, time 7.07s, 
epoch 247, batches 8000|18183, train-acc 0.868, loss 0.351, para-norm 5062.621, grad-norm 10.183, time 7.08s, 
epoch 247, batches 9000|18183, train-acc 0.872, loss 0.339, para-norm 5062.956, grad-norm 5.531, time 7.06s, 
epoch 247, batches 10000|18183, train-acc 0.871, loss 0.346, para-norm 5062.888, grad-norm 3.083, time 7.06s, 
epoch 247, batches 11000|18183, train-acc 0.865, loss 0.356, para-norm 5062.609, grad-norm 5.192, time 7.07s, 
epoch 247, batches 12000|18183, train-acc 0.868, loss 0.348, para-norm 5062.674, grad-norm 11.140, time 7.07s, 
epoch 247, batches 13000|18183, train-acc 0.868, loss 0.352, para-norm 5062.238, grad-norm 6.698, time 7.08s, 
epoch 247, batches 14000|18183, train-acc 0.866, loss 0.350, para-norm 5062.180, grad-norm 9.018, time 7.07s, 
epoch 247, batches 15000|18183, train-acc 0.869, loss 0.346, para-norm 5062.345, grad-norm 7.253, time 7.07s, 
epoch 247, batches 16000|18183, train-acc 0.868, loss 0.348, para-norm 5062.002, grad-norm 16.421, time 7.12s, 
epoch 247, batches 17000|18183, train-acc 0.867, loss 0.353, para-norm 5061.457, grad-norm 4.852, time 7.15s, 
epoch 247, batches 18000|18183, train-acc 0.868, loss 0.345, para-norm 5066.407, grad-norm 7.454, time 7.07s, 
epoch 247, batches 18183|18183, train-acc 0.872, loss 0.346, para-norm 5065.979, grad-norm 1.959, time 1.30s, 
dev-acc 0.862
epoch 248, batches 1000|18183, train-acc 0.871, loss 0.340, para-norm 5066.448, grad-norm 4.630, time 7.18s, 
epoch 248, batches 2000|18183, train-acc 0.870, loss 0.343, para-norm 5066.286, grad-norm 5.752, time 7.08s, 
epoch 248, batches 3000|18183, train-acc 0.871, loss 0.346, para-norm 5066.334, grad-norm 5.138, time 7.08s, 
epoch 248, batches 4000|18183, train-acc 0.866, loss 0.350, para-norm 5065.932, grad-norm 11.812, time 7.07s, 
epoch 248, batches 5000|18183, train-acc 0.869, loss 0.349, para-norm 5065.856, grad-norm 5.982, time 7.08s, 
epoch 248, batches 6000|18183, train-acc 0.871, loss 0.342, para-norm 5065.756, grad-norm 5.866, time 7.08s, 
epoch 248, batches 7000|18183, train-acc 0.872, loss 0.341, para-norm 5065.566, grad-norm 6.526, time 7.07s, 
epoch 248, batches 8000|18183, train-acc 0.868, loss 0.347, para-norm 5065.376, grad-norm 11.756, time 7.08s, 
epoch 248, batches 9000|18183, train-acc 0.868, loss 0.356, para-norm 5065.388, grad-norm 3.322, time 7.09s, 
epoch 248, batches 10000|18183, train-acc 0.870, loss 0.346, para-norm 5065.266, grad-norm 9.201, time 7.08s, 
epoch 248, batches 11000|18183, train-acc 0.867, loss 0.351, para-norm 5064.734, grad-norm 6.537, time 7.09s, 
epoch 248, batches 12000|18183, train-acc 0.870, loss 0.345, para-norm 5064.216, grad-norm 6.789, time 7.09s, 
epoch 248, batches 13000|18183, train-acc 0.865, loss 0.354, para-norm 5063.839, grad-norm 4.326, time 7.07s, 
epoch 248, batches 14000|18183, train-acc 0.866, loss 0.356, para-norm 5063.526, grad-norm 5.503, time 7.08s, 
epoch 248, batches 15000|18183, train-acc 0.866, loss 0.348, para-norm 5062.974, grad-norm 8.558, time 7.09s, 
epoch 248, batches 16000|18183, train-acc 0.869, loss 0.346, para-norm 5062.916, grad-norm 7.010, time 7.19s, 
epoch 248, batches 17000|18183, train-acc 0.867, loss 0.354, para-norm 5062.879, grad-norm 7.666, time 7.07s, 
epoch 248, batches 18000|18183, train-acc 0.866, loss 0.354, para-norm 5062.667, grad-norm 4.337, time 7.07s, 
epoch 248, batches 18183|18183, train-acc 0.864, loss 0.370, para-norm 5062.626, grad-norm 3.497, time 1.30s, 
dev-acc 0.864
epoch 249, batches 1000|18183, train-acc 0.868, loss 0.349, para-norm 5062.858, grad-norm 7.740, time 7.15s, 
epoch 249, batches 2000|18183, train-acc 0.868, loss 0.348, para-norm 5062.962, grad-norm 4.908, time 7.08s, 
epoch 249, batches 3000|18183, train-acc 0.868, loss 0.347, para-norm 5062.568, grad-norm 1.149, time 7.07s, 
epoch 249, batches 4000|18183, train-acc 0.866, loss 0.352, para-norm 5062.672, grad-norm 7.686, time 7.10s, 
epoch 249, batches 5000|18183, train-acc 0.867, loss 0.353, para-norm 5061.814, grad-norm 7.991, time 7.09s, 
epoch 249, batches 6000|18183, train-acc 0.867, loss 0.352, para-norm 5062.184, grad-norm 16.066, time 7.09s, 
epoch 249, batches 7000|18183, train-acc 0.868, loss 0.347, para-norm 5062.493, grad-norm 1.742, time 7.08s, 
epoch 249, batches 8000|18183, train-acc 0.868, loss 0.346, para-norm 5062.067, grad-norm 1.504, time 7.08s, 
epoch 249, batches 9000|18183, train-acc 0.870, loss 0.345, para-norm 5061.807, grad-norm 4.196, time 7.08s, 
epoch 249, batches 10000|18183, train-acc 0.866, loss 0.354, para-norm 5061.712, grad-norm 6.192, time 7.08s, 
epoch 249, batches 11000|18183, train-acc 0.869, loss 0.345, para-norm 5062.369, grad-norm 9.069, time 7.07s, 
epoch 249, batches 12000|18183, train-acc 0.869, loss 0.343, para-norm 5062.500, grad-norm 0.074, time 7.07s, 
epoch 249, batches 13000|18183, train-acc 0.869, loss 0.348, para-norm 5061.731, grad-norm 6.306, time 7.08s, 
epoch 249, batches 14000|18183, train-acc 0.866, loss 0.351, para-norm 5061.145, grad-norm 14.158, time 7.07s, 
epoch 249, batches 15000|18183, train-acc 0.868, loss 0.347, para-norm 5061.369, grad-norm 11.592, time 7.09s, 
epoch 249, batches 16000|18183, train-acc 0.866, loss 0.351, para-norm 5061.113, grad-norm 3.389, time 7.08s, 
epoch 249, batches 17000|18183, train-acc 0.870, loss 0.346, para-norm 5060.873, grad-norm 4.275, time 7.05s, 
epoch 249, batches 18000|18183, train-acc 0.869, loss 0.353, para-norm 5060.763, grad-norm 5.269, time 7.06s, 
epoch 249, batches 18183|18183, train-acc 0.873, loss 0.351, para-norm 5060.764, grad-norm 2.667, time 1.30s, 
dev-acc 0.864
training end!
test-acc 0.862