Skip to content

Fix masking of response tokens #1718

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
merged 1 commit into from
Jun 20, 2024
+2 −6
Merged

Fix masking of response tokens #1718

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
8 changes: 2 additions & 6 deletions trl/trainer/ppo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -614,7 +614,7 @@ def _step_safety_checker(
scores (List[`torch.FloatTensor`]):
List of tensors containing the scores.
masks (List[`torch.LongTensor`], *optional*):
list of optional tensors containing the masks of shape (`query_length` + `response_length`)
list of optional tensors containing the masks of shape (`response_length`)
Returns:
`tuple`: The input processed data.
"""
Expand Down Expand Up @@ -1033,15 +1033,11 @@ def batched_forward_pass(
if attention_mask[j, 0] == 0: # offset left padding
start += attention_mask[j, :].nonzero()[0]
end = start + len(response_batch[j])
if response_masks is not None:
response_masks_batch[j] = torch.cat(
(torch.zeros_like(query_batch[j]), response_masks_batch[j])
)[1:]

masks[j, :start] = 0
masks[j, end:] = 0
if response_masks is not None:
masks[j, start:end] = masks[j, start:end] * response_masks_batch[j][start:end]
masks[j, start:end] = masks[j, start:end] * response_masks_batch[j]

if return_logits:
all_logits.append(logits)
Expand Down