verl icon indicating copy to clipboard operation
verl copied to clipboard

Fix: Pass attn_implementation override to CriticWorker

Open arde171 opened this issue 1 month ago • 0 comments

What does this PR do?

Problem

CriticWorker was extracting `attn_implementation` from `override_config` but not passing it to `load_valuehead_model()`, causing models to always default to `flash_attention_2` regardless of configuration.

Solution

  • Added `attn_implementation` parameter to `load_valuehead_model()` function
  • Implemented smart fallback: explicit param → model_config._attn_implementation → 'flash_attention_2' default
  • Updated CriticWorker to pass `attn_implementation` to `load_valuehead_model()`

Checklist Before Starting

  • [X] Search for similar PRs. Paste at least one query link here: ...
  • [ ] Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
    • {modules} include fsdp, megatron, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data
    • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
    • {type} is in feat, fix, refactor, chore, test
    • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
    • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

[!IMPORTANT] Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

arde171 avatar Nov 25 '25 23:11 arde171