Fix PI0 embed_tokens.weight loading + WandB list metric logging#3116
Open
he-yufeng wants to merge 1 commit intohuggingface:mainfrom
Open
Fix PI0 embed_tokens.weight loading + WandB list metric logging#3116he-yufeng wants to merge 1 commit intohuggingface:mainfrom
he-yufeng wants to merge 1 commit intohuggingface:mainfrom
Conversation
Two fixes: 1. PI0 state dict remapping: handle checkpoints that store embed_tokens.weight directly (not only via lm_head.weight tying). Without this, the key is missing after load_state_dict. 2. WandB wrapper: log list/tuple metrics (like loss_per_dim) as individual per-index entries instead of silently dropping them. Fixes huggingface#3109
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What this PR does
Fixes two issues reported in #3109:
PI0 state dict
embed_tokens.weightmissing after loading:_fix_pytorch_state_dict_keysonly creates this key fromlm_head.weight(weight tying), but some checkpoints storeembed_tokens.weightdirectly. Added explicit handling for this key.WandB silently drops
loss_per_dim: the logging wrapper only handles scalars. List/tuple metrics likeloss_per_dimare now expanded into per-index entries (loss_per_dim_0,loss_per_dim_1, ...).Fixes #3109