[rollout] feat: support async mode for multimodal data inference

[xichengpro]

What does this PR do?

fix #3518

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).)

[gemini-code-assist]

Code Review

I've reviewed your pull request. The main goal is to add support for asynchronous multimodal data inference, which is a great addition. However, I found a critical issue that could lead to a runtime error. Please see my comment below for details and a suggested fix.

[gemini-code-assist]

The code directly accesses image_data['image'] without checking if image_data is None. On line 38, image_data is initialized with kwargs.get("multi_modal_data", None), which means it can be None if multi_modal_data is not in kwargs. This will lead to a TypeError: 'NoneType' is not subscriptable' at runtime.

To prevent this, you should access the 'image' key safely, providing a default value of None if image_data is None or if the key is missing.

Suggested change

	request_id=request_id, prompt_ids=prompt_ids, sampling_params=sampling_params, image_data=image_data['image']
	request_id=request_id, prompt_ids=prompt_ids, sampling_params=sampling_params, image_data=image_data.get("image") if image_data else None

[wuxibin89]

Please also add multi_modal_data to AgentLoopOutput, see: https://github.com/volcengine/verl/blob/main/verl/experimental/agent_loop/tool_agent_loop.py#L171

[xichengpro]

Thank you for the feedback!
I have already added multi_modal_data to AgentLoopOutput.