[perf] feat: support profiler in model engine and sft trainer

[vermouth1992]

What does this PR do?

• As title

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

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 profiling support to the SFT trainer and model engine, which is a great performance analysis feature. The implementation is mostly solid, but I've identified a critical bug in the default profiler configuration that will cause a crash when profiling is enabled. Additionally, there's a significant amount of duplicated code between the standard and Ray-based SFT trainers that should be refactored to improve maintainability. Addressing these points will make the new feature robust and easier to maintain.

[gemini-code-assist]

The step_end value is set to null, which will be parsed as None in Python. This will cause a TypeError in verl/utils/profiler/profile.py during the validation check (self.tool_config.step_end >= 0) and when calculating the active duration for the profiler schedule (self.tool_config.step_end - self.tool_config.step_start).

To fix this, step_end must be an integer. A default value of 1 would be reasonable, allowing the profiler to capture a single step by default.

    step_end: 1

[gemini-code-assist]

The logic for handling the profiler configuration, including the _validate_profiler_interval method, is duplicated from verl/trainer/sft_trainer.py. This code duplication makes the codebase harder to maintain and increases the risk of introducing inconsistencies if one file is updated but the other is not.

Consider refactoring the common logic from SFTTrainer in both sft_trainer.py and sft_trainer_ray.py into a shared base class. This would centralize the configuration handling and other shared functionalities, leading to cleaner and more maintainable code.