Fix grpo nan #3278
Closed
Fix grpo nan #3278
+133
−58
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
unsloth/models/rl.py
Outdated
| # Selective log softmax | ||
| selective_log_softmax_code = inspect.getsource(selective_log_softmax) | ||
|
|
||
| #GRPO masking code |
There was a problem hiding this comment.
Suggested change
| #GRPO masking code | |
| # GRPO masking code |
unsloth/models/rl_replacements.py
Outdated
|
|
||
| # The new lines you want to insert | ||
| replacement_lines = """batch_size = self.args.per_device_train_batch_size if mode == "train" else self.args.per_device_eval_batch_size | ||
| prompt_completion_ids = left_pack_padding(prompt_completion_ids, self.processing_class.pad_token_id)""" |
There was a problem hiding this comment.
In order to easierly resolve merge conflicts when testing with the Fast VLM infernece branch I moved everything in this PR to: #3132
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
Fixes the grpo nan issues we have been having with ga steps > 1, tested on h100 and collab on T4. This PR was created mainly to avoid passing a SPDA attention mask so it would not eat up a lot of memory. Relies on unslothai/unsloth-zoo#265.