[megatron] chore: clean legacy code path part 1, make engine use mbridge by default

[ISEEKYAN]

What does this PR do?

this is one of a series PRs to clean the legacy megatron code path and make bridge the default path for megatron. #4496

This PR make megatron engine use only support bridge, and mark the legacy code in mcore/registry.py

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

This pull request is the first part of a series to clean up legacy code paths in Megatron and make mbridge the default. The changes enforce the use of mbridge in the Megatron engine and remove the old code paths that did not use it. The logic for selecting forward functions has also been refactored.

My review found two critical issues in verl/models/mcore/registry.py where checking for Vision Language Model (VLM) architectures is implemented incorrectly. The current implementation will fail to identify VLMs, causing incorrect forward functions to be used. I have provided suggestions to fix this.

[gemini-code-assist]

The check hf_config.architectures[0] in SupportedVLM is incorrect. The in operator on an Enum checks for member identity, not the string value of the members. Since hf_config.architectures[0] is a string, this condition will always evaluate to False, leading to the wrong forward function being selected for Vision Language Models (VLMs).

Suggested change

	if hf_config.architectures[0] in SupportedVLM:
	if hf_config.architectures[0] in {item.value for item in SupportedVLM}:

[gemini-code-assist]

Similar to the issue in get_mcore_forward_fn, the check hf_config.architectures[0] in SupportedVLM is incorrect. The in operator on an Enum checks for member identity, not the string value of the members. This will always evaluate to False, causing the wrong fused forward function to be selected for Vision Language Models (VLMs).

Suggested change

	if hf_config.architectures[0] in SupportedVLM:
	if hf_config.architectures[0] in {item.value for item in SupportedVLM}:

[wuxibin89]

We should also set use_mbridge to True? https://github.com/volcengine/verl/blob/main/verl/trainer/config/engine/megatron.yaml#L77-L78