SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fix steps bugs in no trainer examples (#24197)

Browse Source 
Fix step bugs in no trainer + load checkpoint + grad acc
This commit is contained in:
Ethan
2023-06-12 23:49:55 +08:00
committed by GitHub
parent 08ae37c820
commit f7d80cb3d2
12 changed files with 34 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
examples/pytorch/multiple-choice/run_swag_no_trainer.py
View File

@@ -559,10 +559,11 @@ def main():
resume_step = None
completed_steps = starting_epoch * num_update_steps_per_epoch
else:
resume_step = int(training_difference.replace("step_", ""))
# need to multiply `gradient_accumulation_steps` to reflect real steps
resume_step = int(training_difference.replace("step_", "")) * args.gradient_accumulation_steps
starting_epoch = resume_step // len(train_dataloader)
resume_step -= starting_epoch * len(train_dataloader)
completed_steps = resume_step
completed_steps = resume_step // args.gradient_accumulation_stepp
# update the progress_bar if load from checkpoint
progress_bar.update(completed_steps)