[ci] feat: add profiling tests to vLLM ci

[Gary-cjy]

What does this PR do?

This PR integrates NPU profiling capabilities into the vLLM CI pipeline. It enables detailed performance monitoring for both the Actor and Reference components within the actor_rollout_ref module. By adding these profiling tests, we can capture execution traces, analyze hardware utilization (NPU/CPU), and identify performance bottlenecks during the CI process.

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, cfg, reward
  • 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

The changes have been tested by running the updated CI script on an Ascend NPU environment.

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).)
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

[CLAassistant]

All committers have signed the CLA.

[gemini-code-assist]

Code Review

This pull request adds a shell script for a profiling test. I've found two critical issues in the script that will cause it to fail. One is the use of a placeholder path for saving results, which is not suitable for a CI environment. The other is a shell syntax error in a variable assignment. Both issues need to be fixed for the script to run correctly.

[gemini-code-assist]

The SAVE_PATH is set to a placeholder "your_path". This is not a valid directory and will cause the script to fail, especially in a CI environment where such a path is unlikely to exist. Please use a valid, descriptive path for the output data.

Suggested change

	SAVE_PATH="your_path"
	SAVE_PATH="outputs/profile_qwen2_5_05b_grpo"

[gemini-code-assist]

The assignment CONTENTS=['npu','cpu'] is invalid shell syntax. The shell will attempt to execute ['npu','cpu'] as a command and assign its output to the CONTENTS variable, which will fail. To assign the string representation of a list to the variable, the value must be enclosed in quotes.

Suggested change

	CONTENTS=['npu','cpu']
	CONTENTS="['npu','cpu']"

[tardis-key]

trainer.device is not required, device type can be obtained automatically