[trainer] fix: avoid loading duplicated custom reward function to fix issue #3150

[fshp971]

What does this PR do?

This PR attempts to fix issue #3150, where custom reward function cannot be used with reward_model.reward_manager=prime.

The main cause of issue #3150 is that when using verl.trainer.ppo.reward.get_custom_reward_fn to create reward_fn and val_reward_fn for the RayPPOTrainer, the custom reward code file is loaded twice under the same module name "custom_module".

This results in the actual reward functions reward_fn.compute_score and val_reward_fn.compute_score having the same module-function name (i.e., "custom_module") but different id() values in memory, and only val_reward_fn.compute_score has the correct, latest id() address.

Therefore, when using the "prime" reward manager, since it tries to use concurrent.futures.ProcessPoolExecutor to pickle the reward functions for multiprocessing, the ProcessPoolExecutor finds that reward_fn.compute_score has an incorrect id() compared to its global "custom_module"'s recorded id(), and thus fails to pickle reward_fn.compute_score.

The solution for this bug is to simply avoid loading duplicate modules from the same custom file in verl.trainer.ppo.reward.get_custom_reward_fn.

Checklist Before Starting

• 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.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[gemini-code-assist]

Code Review

This pull request correctly addresses issue #3150 by preventing the duplicate loading of a custom reward function module. The change checks sys.modules before importing the module from a file, which resolves pickling errors with ProcessPoolExecutor. I've added one suggestion to further improve the robustness of this fix by ensuring the specified reward function is validated in the module, whether it's newly loaded or was pre-existing.

[vermouth1992]

Could you comment on Gemini's review?

[fshp971]

Hi,

I think the suggestion made by Gemini would make the code more robust.
I have committed that.