Feat/log fit sub reward

[Arinlzy]

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.

• Add the logging mechanism for extra sub-reward in training/fitting phase
• related issue: 【提问】如何在verl中设置观察 每种reward的变化曲线图呢？ #2279

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...(no similar PR found..)
• 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).)

[gemini-code-assist]

Code Review

This pull request adds a logging mechanism for extra sub-rewards during the training phase. A new static method _summarize_reward_extras is introduced to process and summarize this information into metrics. The changes are well-integrated into both synchronous and asynchronous reward computation paths.

My review focuses on the implementation of the new _summarize_reward_extras method. I've identified a potential issue where the data flattening and conversion logic could silently ignore some data formats (like multi-element tensors/arrays or deeply nested lists), leading to incorrect metrics. I've provided a suggestion to make this logic more robust.

[gemini-code-assist]

The current implementation for processing reward_extra_infos is not fully robust. It only flattens a single level of nested list or tuple objects and silently ignores multi-element torch.Tensor or np.ndarray objects. This can lead to incomplete data and misleading metrics.

For example, if a value is np.array([1, 2]) or a nested list like [[1, 2]], its contents will not be included in the statistics.

A more robust approach would be to recursively flatten all iterable structures and then attempt to convert all resulting items to floats. This ensures all numeric data is captured, regardless of nesting or type.

            # 1. Flatten all nested lists, tuples, tensors, and arrays
            flattened: list = []
            items_to_flatten = list(values)
            while items_to_flatten:
                item = items_to_flatten.pop(0)
                if isinstance(item, (list, tuple)):
                    items_to_flatten.extend(item)
                elif isinstance(item, (torch.Tensor, np.ndarray)):
                    items_to_flatten.extend(item.flatten().tolist())
                else:
                    flattened.append(item)

            # 2. Convert flattened items to numeric values
            numeric_vals: list[float] = []
            for value in flattened:
                try:
                    numeric_vals.append(float(value))
                except (TypeError, ValueError):
                    # Ignore values that cannot be converted to float
                    pass