[Tests] Modify test cases

[amy-why-3459]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

In the VLLM implementation, if distributed_executor_backend is not configured, it will be selected based on world_size.

if self.distributed_executor_backend is None and self.world_size_across_dp > 1:
            # We use multiprocessing by default if world_size fits on the
            # current node and we aren't in a ray placement group.

            from vllm.v1.executor import ray_utils

            backend: DistributedExecutorBackend = "mp"
            ray_found = ray_utils.ray_is_available()
            if current_platform.is_tpu() and envs.VLLM_XLA_USE_SPMD:
                backend = "uni"
            elif current_platform.is_cuda() and self.nnodes > 1:
                backend = "mp"
            elif (
                current_platform.is_cuda()
                and current_platform.device_count() < self.world_size
            ):
                gpu_count = current_platform.device_count()
                raise ValueError(
                    f"World size ({self.world_size}) is larger than the number of "
                    f"available GPUs ({gpu_count}) in this node. If this is "
                    "intentional and you are using:\n"
                    "- ray, set '--distributed-executor-backend ray'.\n"
                    "- multiprocessing, set '--nnodes' appropriately."
                )
            elif self.data_parallel_backend == "ray":
                logger.info(
                    "Using ray distributed inference because "
                    "data_parallel_backend is ray"
                )
                backend = "ray"
            elif ray_found:
                if self.placement_group:
                    backend = "ray"
                else:
                    from ray import is_initialized as ray_is_initialized

                    if ray_is_initialized():
                        from ray.util import get_current_placement_group

                        if get_current_placement_group():
                            backend = "ray"
            self.distributed_executor_backend = backend
            logger.debug("Defaulting to use %s for distributed inference", backend)

        if self.distributed_executor_backend is None and self.world_size == 1:
            self.distributed_executor_backend = "uni"

Test Plan

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (Optional) Release notes update. If your change is user-facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[hsliuustc0106]

what's the maximal in theory?

[amy-why-3459]

what's the maximal in theory?

In async_chunk mode, since chunked prefill is not currently supported, the theoretical maximum supported concurrency is 26 (65536/2500) for an input of 2500.

[gcanlin]

Could you help check whether this default mp distributed_executor_backend will be applied in every stage? I notice that if so, it's making UX degradation. Because vLLM has a complete init for distributed_executor_backend. How about removing it and make every stage choose by themselves even when default? cc @lishunyang12

https://github.com/vllm-project/vllm/blob/7b1e0b07d0ea9fe11f0c4a35001baade2c3b1074/vllm/config/parallel.py#L790-L839

[amy-why-3459]

Could you help check whether this default mp distributed_executor_backend will be applied in every stage? I notice that if so, it's making UX degradation. Because vLLM has a complete init for distributed_executor_backend. How about removing it and make every stage choose by themselves even when default? cc @lishunyang12

https://github.com/vllm-project/vllm/blob/7b1e0b07d0ea9fe11f0c4a35001baade2c3b1074/vllm/config/parallel.py#L790-L839

I completely agree with your point. I think we can discuss whether we need to set a default value for the distributed_executor_backend parameter.

[hsliuustc0106]

BLOCKING:

• Breaking Changes — This PR changes the default distributed_executor_backend from "mp" to "uni" in vllm_omni/config/stage_config.py:422. This is a breaking change that affects all deployments using the default DeployConfig. Please:

  1. Split this PR into two: one for the test changes, one for the config default change
  2. For the config default change PR, add justification for why "uni" should be the new default
  3. Add a migration guide or note in the PR description about how existing deployments will be affected

• Documentation — The PR description doesn't mention the config default change, making it unclear that this is a breaking change. Please update the PR title (e.g., [Breaking Change]) and body to document this change.

[amy-why-3459]

BLOCKING:

• Breaking Changes — This PR changes the default distributed_executor_backend from "mp" to "uni" in vllm_omni/config/stage_config.py:422. This is a breaking change that affects all deployments using the default DeployConfig. Please:

  1. Split this PR into two: one for the test changes, one for the config default change
  2. For the config default change PR, add justification for why "uni" should be the new default
  3. Add a migration guide or note in the PR description about how existing deployments will be affected

• Documentation — The PR description doesn't mention the config default change, making it unclear that this is a breaking change. Please update the PR title (e.g., [Breaking Change]) and body to document this change.

I will submit a separate PR to remove the default values ​​from the config file.

[amy-why-3459]

@gcanlin @hsliuustc0106 I think this PR is ready and can be merged.

[gcanlin]

LGTM