bugfix:Fix mbridge weight save logic#119
Open
zyang6 wants to merge 1 commit into
Open
Conversation
4 tasks
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Model Issue Type: DeepSeek-like model with MTP modules
Symptom: Incomplete weight saving
Root Cause Analysis: The model contains MTP layers during inference, yet these layers become inactive and do not participate in reinforcement learning training. As a result, corresponding weights for the MTP layers are absent.However, the mbridge saving mode enforces strict key-value integrity verification for all entries inside the .safetensors file, requiring full exact matching.When MTP layers are excluded from training, weights of the 61st layer get lost, which further causes the final .safetensors file to fail saving completely.
Solution Idea: Check whether incomplete safetensors weights cause other model parameters to fail saving in the final fallback procedure.Ensure such remaining weights are saved normally. Meanwhile, output and log the missing weight keys in the safetensors file to prompt users.