[Feature] Add CI test for weight version update#1120
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| def update_weights(self) -> None: | ||
| self.weight_version += 1 |
There was a problem hiding this comment.
UpdateWeight.update_weights now increments self.weight_version (lines 46‑47), but UpdateWeightFromDistributed overrides __init__ without calling the base constructor, so instances created when args.colocate is False never define weight_version. The next call to update_weights() raises AttributeError before any tensors are sent, breaking non-colocated training runs. Make sure UpdateWeightFromDistributed sets self.weight_version (e.g., by invoking super().init) so distributed weight sync can run.
Useful? React with 👍 / 👎.
2 participants
No description provided.