Ensure Equivalence With DPO Padding and Padding Free #34
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
garrett361
reviewed
Jul 23, 2025
open_instruct/dpo_tune_cache.py
Outdated
| simpo_loss, | ||
| wpo_loss, | ||
| ) | ||
| from open_instruct.padding_free_collator import ( |
Collaborator
Author
There was a problem hiding this comment.
yes there is a dup two lines belwo
garrett361
reviewed
Jul 23, 2025
open_instruct/dpo_utils.py
Outdated
| if not packing: | ||
| concatenated_batch = concatenated_inputs(batch) | ||
| try: | ||
| pad_token_id = model.tokenizer.pad_token_id |
Owner
There was a problem hiding this comment.
Do models usually have a tokenizer attr? I'm finding not, for bamba and granite. Unless the attr is set somewhere after init.
Collaborator
Author
There was a problem hiding this comment.
they will usually have, though the model is typed as torch.nn.Module so I thought in the OI use-case it may not be gauranteed.
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
This reverts commit 266f214.
garrett361
reviewed
Jul 24, 2025
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
garrett361
reviewed
Jul 24, 2025
Owner
|
also, quality checks failing |
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
Collaborator
Author
|
@garrett361 fixed conflict and linted the files effectively chaged in this commit. But because we had to pull upstream, there are alot of files not linted |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Actually I believe that padding-free (packing) is correct.
this PR fixes the non-packing case, whereconcatenated_forwardassumes thepad_token_id = 0by mistakeseq_idxin the padding freeconcatenated_forwardwas computed wrongly.preference_span_searchfunction.Checking