[ci] feat: add GRPO Qwen3.5-35B Megatron vLLM nightly test for Ascend

[chengminhua]

What does this PR do?

Add a new nightly CI workflow for GRPO Qwen3.5-35B Megatron training on Ascend using the vLLM rollout backend.

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, 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 a new bash script run_grpo_qwen3_5_35b_megatron_npu.sh to run GRPO training with Qwen3.5-35B using Megatron on NPU. Feedback highlights two critical issues: first, the expert model parallel size (EP=8) is invalid for the calculated data parallel size (DP=4), which will trigger an assertion error; second, hardcoding /root as the log directory can lead to permission issues in non-root environments.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

With n_devices_per_node=16 and trainer.nnodes=1, the total number of GPUs is 16. Given TP=2, PP=2, and CP=1, the data parallel size (DP) is calculated as:

DP = world_size / (TP * PP * CP) = 16 / (2 * 2 * 1) = 4.

In Megatron Core, the expert model parallel size (EP) must divide the data parallel size (DP). Since DP = 4, setting EP = 8 is invalid because 4 is not divisible by 8 (and EP cannot exceed DP). This will cause an assertion error at startup:
AssertionError: data_parallel_size (4) must be divisible by expert_model_parallel_size (8).

To fix this, EP should be set to 4 or less.

Suggested change

	EP=${EP:-8}
	EP=${EP:-4}

[gemini-code-assist]

Hardcoding /root as the log directory can cause permission denied errors if the script is run in a non-root environment (e.g., standard CI runners or local development environments). It is safer and more consistent with MODEL_PATH (on line 6) to use ${HOME} instead.

Suggested change

	LOG_DIR=/root/.cache/nightly_log/$SCRIPT_NAME
	LOG_DIR=${HOME}/.cache/nightly_log/$SCRIPT_NAME