[megatron] feat: LoRA adapter only refit (TensorLoRARequest)

[HollowMan6]

What does this PR do?

Waiting for:

• [Draft] Peft Bridge NVIDIA-NeMo/Megatron-Bridge#1766
• [for #1766] fix MoE expert layers LoRA bridge NVIDIA-NeMo/Megatron-Bridge#1788
• [BugFix] LoRA: Support loading base_layer of experts vllm-project/vllm#31104

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 introduces a significant feature for LoRA adapter-only weight updates in the Megatron backend, which avoids the overhead of merging adapters into the base model for each weight synchronization. The changes are well-structured and include updates to documentation, configuration handling for LoRA, new PEFT utility functions for vLLM compatibility, and refactored weight export logic. The implementation appears solid and aligns with the stated goals. I have a couple of suggestions to enhance code maintainability and clarity regarding duplicated logic and a confusing condition.

[gemini-code-assist]

Code Review

This pull request introduces a valuable feature for LoRA adapter-only refitting in the Megatron backend, which should provide significant performance benefits. The changes across documentation, configuration, and worker implementations appear to correctly support both merging LoRA adapters and loading them separately. My primary concern, detailed in a specific comment, is the repeated implementation of LoRA configuration logic across several files. Addressing this will improve the long-term maintainability of the codebase.