Do not disable weight tying on meta-init ranks - #856
Conversation
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Merge Protections🔴 1 of 1 protections blocking · waiting on 👀 reviews
🔴 Require one maintainer reviewWaiting for any of
This rule is failing.All PRs must have at least one approving review from a maintainer before merging.
|
Non-source ranks load on the `meta` device and previously set `tie_word_embeddings=False`. For models with tied word embeddings, that turns the tied output projection (e.g. `lm_head.weight`) into a standalone parameter absent from the checkpoint, so it is reported as a missing key and never populated. Remove the default so weights are tied consistently across ranks. Add `test_load_no_missing_keys`, which loads a range of models (multimodal, tied, and untied) under `load_offloaded_model` and asserts that transformers reports no missing keys. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
78874c7 to
31c7e79
Compare
Purpose
Changes
tied_embeddings=Falseoverride when loading non-source ranksTesting