feat:support exporting fp8 weights in FP8 E2E

[eternally-z]

What does this PR do?

we support exporting fp8 weights in Megatron-Bridge. This pr enable verl using the feature.

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, vllm_omni, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data, cfg, reward, fully_async, one_step_off
  • 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.

[gemini-code-assist]

Code Review

This pull request introduces support for FP8 weight export and rollout by adding export_weight_dtype and skip_mid_quantization configurations. It updates the Megatron engine and workers to handle FP8 weight initialization and optimizer parameter reloading, and ensures the Prometheus metrics directory is created on Ray workers. Feedback was provided regarding redundant validation and normalization of the export weight data type in the optimizer initialization logic, suggesting the use of the already validated value stored in the bridge.

[gemini-code-assist]

The normalization and validation of export_weight_dtype is redundant here. It has already been processed and validated in _build_tf_config (lines 238-245) and stored in self.bridge.export_weight_dtype. Repeating this logic here is redundant and could lead to inconsistencies if the validation logic changes in one place but not the other.

Suggested change

	export_weight_dtype = self.engine_config.export_weight_dtype
	if isinstance(export_weight_dtype, str):
	export_weight_dtype = export_weight_dtype.lower()
	if export_weight_dtype == "fp8" and self.bridge is not None:
	if self.bridge is not None and getattr(self.bridge, "export_weight_dtype", None) == "fp8":

[gemini-code-assist]

This block redundantly re-fetches and re-normalizes export_weight_dtype from the configuration. This logic is already handled in _init_hf_config_and_tf_config (lines 239-246), where the value is validated (raising ValueError for invalid inputs) and stored in self.bridge.export_weight_dtype. Additionally, the fallback to "bf16" at line 534 is inconsistent with the ValueError raised earlier in the initialization flow. Using the stored value from the bridge is safer and cleaner.

Suggested change

	export_weight_dtype = self.config.actor.megatron.get("export_weight_dtype", "bf16")
	if isinstance(export_weight_dtype, str):
	export_weight_dtype = export_weight_dtype.lower()
	if export_weight_dtype not in ("bf16", "fp16", "fp8"):
	export_weight_dtype = "bf16"
	if export_weight_dtype == "fp8" and self.bridge is not None:
	if self.bridge is not None and getattr(self.bridge, "export_weight_dtype", None) == "fp8":