-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.sh
31 lines (30 loc) · 1.62 KB
/
train.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
data_dir=./data/depth-5
lr=1e-5
ep=50
output_dir=./output/d5
# model=roberta_rr # PRover
model=roberta_probr # PRobr
num_gpus=8
distribute_on=" -m torch.distributed.launch --nproc_per_node=$num_gpus"
distribute_off=""
distribute=$distribute_off
python3 $distribute ./run_experiment_probr.py \
--data_dir $data_dir \
--output_dir $output_dir \
--per_gpu_eval_batch_size 32 \
--per_gpu_train_batch_size 8 \
--model_type $model \
--model_name_or_path roberta-large \
--task_name rr \
--do_train \
--do_eval \
--do_lower_case \
--max_seq_length 300 \
--learning_rate $lr \
--gradient_accumulation_steps 1 \
--num_train_epochs $ep \
--logging_steps 4752 \
--save_steps 4750 \
--seed 42 \
--data_cache_dir ./output/cache/ \
--evaluate_during_training \