[perf] fix: infer server profiler args fix

[mengchengTang]

What does this PR do?

Description

This PR addresses issues with profiler configuration for inference engines (vLLM and SGLang) and refactors the profiler code for better cohesion and maintainability.

Key Changes:

• vLLM Profiler Configuration Fix:
  Implemented automatic setup of profiler-related environment variables for vLLM engine before process launch. Users no longer need to manually set vLLM environment variables.
  Added support for configuring the profiler via profiler_config for newer vLLM versions, ensuring compatibility.

• SGLang Profiler Bug Fix:
  Fixed a bug introduced in @PR 5025 regarding SGLang profiler data collection.
  Corrected the argument construction logic to ensure accurate profiling parameters are passed to the SGLang backend.

• Code Refactoring & Cohesion:
  Centralized profiler logic by introducing adapter functions (build_vllm_profiler_args, build_sglang_profiler_args) within the profiler component.
  These adapters encapsulate the logic for constructing engine-specific profiler arguments, improving code organization and readability.

• Testing:
  Added comprehensive unit tests to verify the correctness of the new adapter functions and ensure regression testing for the fixed bugs.

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, veomni, 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

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).)
• 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.

[wuxibin89]

@mengchengTang Please fix sanity check.

[mengchengTang]

@mengchengTang Please fix sanity check.

Got it, I'll fix it soon and add GPU/NPU verification results.

---

2026/02/04 update gpu and npu test result：

Scene	Discrete Mode	With Stack	with_memory	Record Shapes	Steps	Ranks	Status
vLLM 0.11.0 + NPU	TRUE	FALSE	FALSE	FALSE	[2]	[1]	pass
vLLM 0.11.0 + NPU	TRUE	TRUE	TRUE	TRUE	[2]	All	pass
vLLM 0.11.0 + GPU	TRUE	FALSE	FALSE	FALSE	[2]	[1]	pass
vLLM 0.11.0 + GPU	TRUE	TRUE	TRUE	TRUE	[2]	All	pass
vLLM 0.13.0 + NPU	TRUE	FALSE	FALSE	FALSE	[2]	[1]	pass
vLLM 0.13.0 + NPU	TRUE	TRUE	TRUE	TRUE	[2]	All	pass
vLLM 0.13.0 + GPU	TRUE	FALSE	FALSE	FALSE	[2]	[1]	pass
vLLM 0.13.0 + GPU	TRUE	TRUE	TRUE	TRUE	[2]	All	pass
SGLang + NPU	TRUE	FALSE	FALSE	FALSE	[2]	[1]	pass
SGLang + NPU	TRUE	TRUE	TRUE	TRUE	[2]	All	pass
SGLang + GPU	TRUE	FALSE	FALSE	FALSE	[2]	[1]	pass
SGLang + GPU	TRUE	TRUE	TRUE	TRUE	[2]	All	pass

[tardis-key]

nice