[xiaohong42]

What does this PR do?

Removes the vLLM check that verl-project#8371 added to examples/grpo_trainer/run_deepseek_v4_veomni.sh. Where the recipe turns on AITER (gfx94x), it exits unless the installed vLLM defines VLLM_ROCM_DSV4_BPRESHUFFLE. That variable was proposed in vllm-project/vllm#63604, which was closed without being merged, so no vLLM defines it and the recipe refuses every vLLM on gfx94x:

With AITER this vLLM B-preshuffles DeepSeek-V4's FP8 weights, so the rollouts would run on corrupted weights after the first weight sync. Use a vLLM with VLLM_ROCM_DSV4_BPRESHUFFLE (vllm-project/vllm#63604).

Redoing the B-preshuffle when weights are reloaded belongs in vLLM, so the recipe no longer inspects vLLM for it. The requirement in the recipe header goes with the check.

Checklist Before Starting

• Search for similar PRs: https://github.com/verl-project/verl/pulls?q=VLLM_ROCM_DSV4_BPRESHUFFLE
• Format the PR title as [{modules}] {type}: {description}

Test

On gfx942 with a vLLM that does not define VLLM_ROCM_DSV4_BPRESHUFFLE, with the trainer launch replaced by an echo: on main the recipe exits with the message above (exit code 1); with this PR it reaches the launch and still passes NCCL_DEBUG=VERSION and VLLM_ROCM_USE_AITER=1 to the workers. With VLLM_ROCM_USE_AITER=0, where the check does not run, main and this PR print the same launch arguments. bash -n passes.

API and Usage Example

No API or config change.

Design & Code Changes

• examples/grpo_trainer/run_deepseek_v4_veomni.sh: remove the vLLM check and its requirement in the header. The worker environment block is unchanged.

Not changed here: verl/plugin/platform/platform_rocm.py still defaults VLLM_ROCM_DSV4_BPRESHUFFLE=0 for the rollout servers. No vLLM reads it, so it has no effect.

Checklist Before Submitting

• Read the Contribute Guide.
• Apply pre-commit checks: shell only, bash -n; check_example_naming.py and check_uv_gpu_only.py pass.
• Add / Update the documentation: not needed beyond the recipe header.
• Add unit or end-to-end test(s): the recipe test that covered this check was not kept when [rollout, hardware, recipe] fix: run the DeepSeek-V4 VeOmni recipe on AMD ROCm out of the box verl-project/verl#8371 merged; checked by hand as above.
• Once your PR is ready for CI, send a message in the ci-request channel of the verl Slack workspace.
• recipe submodule: not changed.