[recipe] feat: modify dapo deepseek megatron script

[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
  • 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 updates hyperparameters for a DAPO training script. The changes appear to be aimed at tuning the experiment, for instance by increasing the total number of responses generated per training step. My review identified that one of the parameter changes also corrects a latent bug in the previous configuration related to Megatron's pipeline parallelism constraints. I've provided a high-severity comment to improve the clarity of a comment associated with this parameter to prevent future misconfigurations.

[gemini-code-assist]

The comment on this line is ambiguous and does not accurately reflect the underlying constraint for Megatron's pipeline parallelism. An incorrect or unclear comment on such a critical parameter can easily lead to invalid configurations and training failures in the future.

The actual constraint is that the number of micro-batches must be a multiple of the pipeline parallel size (pp). This change from 16 to 32 correctly sets the number of micro-batches to 16 (32 / 2), which is a multiple of pp=16, fixing a latent issue from the previous configuration where the number of micro-batches was 8.

To improve clarity and prevent future errors, I recommend updating the comment to state the correct constraint explicitly.

Suggested change

	train_prompt_mini_bsz=32 # mini_bsz * n >= micro_bsz * pp * dp
	train_prompt_mini_bsz=32 # num_microbatches (mini_bsz/micro_bsz_per_gpu) must be a multiple of pp. (32/2=16, pp=16)

[ccclyu]

thanks! Let us wait to merge until I finish the validation and revise some configs accordingly.