problem

[HeliaHashemipour]

Hi,
Can you tell me what is
train_num_steps and timesteps?
I can't get the difference between them. (which one of them is the timesteps in the paper?)

[dancingacidpanda]

train_num_steps is the number of times the gradient is computed (for every train_batch_size * gradient_accumulate_every samples in your dataset) until the model stops training

timesteps is "the one from the paper"