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

Update no_trainer script for summarization (#19277)

Browse Source 
* Update no_trainer script for summarization

* removed unnecessary import

* fixes notation mistake

* removed: unused variable
This commit is contained in:
Divyanshu Kumar
2022-10-03 18:51:51 +05:30
committed by GitHub
parent 36f52e9593
commit c28d04e9e2
1 changed files with 2 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
examples/pytorch/summarization/run_summarization_no_trainer.py
View File

@@ -669,7 +669,6 @@ def main():
"max_length": args.val_max_target_length if args is not None else config.max_length, "max_length": args.val_max_target_length if args is not None else config.max_length,
"num_beams": args.num_beams, "num_beams": args.num_beams,
} }
samples_seen = 0
for step, batch in enumerate(eval_dataloader): for step, batch in enumerate(eval_dataloader):
with torch.no_grad(): with torch.no_grad():
generated_tokens = accelerator.unwrap_model(model).generate( generated_tokens = accelerator.unwrap_model(model).generate(
@@ -686,7 +685,7 @@ def main():
# If we did not pad to max length, we need to pad the labels too # If we did not pad to max length, we need to pad the labels too
labels = accelerator.pad_across_processes(batch["labels"], dim=1, pad_index=tokenizer.pad_token_id) labels = accelerator.pad_across_processes(batch["labels"], dim=1, pad_index=tokenizer.pad_token_id)
generated_tokens, labels = accelerator.gather((generated_tokens, labels)) generated_tokens, labels = accelerator.gather_for_metrics(generated_tokens, labels)
generated_tokens = generated_tokens.cpu().numpy() generated_tokens = generated_tokens.cpu().numpy()
labels = labels.cpu().numpy() labels = labels.cpu().numpy()
@@ -699,14 +698,8 @@ def main():
decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True) decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels) decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
# If we are in a multiprocess environment, the last batch has duplicates
if accelerator.num_processes > 1:
if step == len(eval_dataloader) - 1:
decoded_preds = decoded_preds[: len(eval_dataloader.dataset) - samples_seen]
decoded_labels = decoded_labels[: len(eval_dataloader.dataset) - samples_seen]
else:
samples_seen += len(decoded_labels)
decoded_preds, decoded_labels = accelerator.gather_for_metrics(decoded_preds, decoded_labels)
metric.add_batch( metric.add_batch(
predictions=decoded_preds, predictions=decoded_preds,
references=decoded_labels, references=decoded_labels,