verl icon indicating copy to clipboard operation
verl copied to clipboard

[fsdp] fix: Fixing the error caused by empty tensors in the multi_turn + remove_padding scenario

Open nuerxiati opened this issue 1 month ago • 2 comments

fix the BUG caused by empty tensors in the multi_turn + remove_padding scenario

What does this PR do?

Add concise overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.

In multi-turn scenarios, the attention_mask may encounter full masking. When use_remove_padding is enabled and mbs is sufficiently small, the forward input_ids and position_ids may become an empty tensor, triggering an error.

This PR creates an input containing at least one element during full masking and removes it from the output, preventing this error and enhancing code robustness.

Checklist Before Starting

  • [x] Search for similar PRs. Paste at least one query link here: ...
  • [x] 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.

I used qwen 3b with mbs set to 1, and compared the reward between enabling use_remove_padding (the branch will follow the modified code) and not enabling it. image

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.

nuerxiati avatar Nov 17 '25 09:11 nuerxiati

@FightingZhen @ji-huazhong,please help me review

nuerxiati avatar Nov 19 '25 02:11 nuerxiati

CLA assistant check
All committers have signed the CLA.

CLAassistant avatar Dec 02 '25 13:12 CLAassistant