Skip to content

Fix GPT_neox incorrect output with batch query #1358

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
regisss merged 1 commit into from
Oct 16, 2024
Merged

Fix GPT_neox incorrect output with batch query #1358

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions optimum/habana/transformers/generation/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2437,6 +2437,17 @@ def _sample(
if streamer is not None:
streamer.end()

if batch_size > 1 and has_eos_stopping_criteria:
eos_token_id = generation_config.eos_token_id
idx_bs = generation_config.max_length
for i in range(batch_size):
for idx in range(len(input_ids[i])):
if input_ids[i][idx] == eos_token_id:
idx_bs = idx
if idx > idx_bs:
input_ids[i][idx] = pad_token_id
idx_bs = generation_config.max_length
Copy link
Copy Markdown
Collaborator

@libinta libinta Oct 14, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

do we still want to return padding token in this case or just stop here?

Copy link
Copy Markdown
Contributor Author

@Jianhong-Zhang Jianhong-Zhang Oct 14, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

according to customer requirement, we need to padding token after <|endoftext|>

Copy link
Copy Markdown

@bhargaveede bhargaveede Nov 21, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

https://huggingface.co/facebook/bart-large-cnn/blob/main/config.json
For bart decoder_start_token_id and eos_token_id are same.
This breaks bart generation


if return_dict_in_generate:
if self.config.is_encoder_decoder:
return GenerateEncoderDecoderOutput(
Expand Down