Disable spec-decode + chunked-prefill for draft models with tensor parallelism > 1

[sroy745]

In this pr we are disabling the simulataneous enablement of both spec-decode & chunked-prefill if the draft model has a tp of > 1. This is needed because as reported in #9291 we see errors if both are enabled for draft model with tp> 1. After this change for this command we return the following error

python3 -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --port 8083 --model "meta-llama/Meta-Llama-3.1-8B-Instruct"   --speculative_model meta-llama/Llama-3.1-8B-Instruct   --tensor-parallel-size 4    --disable-log-requests --use-v2-block-manager --seed 42 --num_speculative_tokens 5  --spec-decoding-acceptance-method typical_acceptance_sampler  --speculative_draft_tensor_parallel_size 4 --max-num-seqs 64 --enable-chunked-prefill


INFO 11-08 03:44:35 api_server.py:166] Multiprocessing frontend to use ipc:///tmp/052b29fd-c3d4-462e-a184-4bfefd9cc94a for IPC Path.
INFO 11-08 03:44:35 api_server.py:185] Started engine process with PID 1182727
INFO 11-08 03:44:40 config.py:347] This model supports multiple tasks: {'generate', 'embedding'}. Defaulting to 'generate'.
INFO 11-08 03:44:40 config.py:1014] Defaulting to use mp for distributed inference
Traceback (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/home/jovyan/cp-sd-check/vllm/entrypoints/openai/api_server.py", line 615, in <module>
    uvloop.run(run_server(args))
  File "/usr/local/lib/python3.10/dist-packages/uvloop/__init__.py", line 82, in run
    return loop.run_until_complete(wrapper())
  File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
  File "/usr/local/lib/python3.10/dist-packages/uvloop/__init__.py", line 61, in wrapper
    return await main
  File "/home/jovyan/cp-sd-check/vllm/entrypoints/openai/api_server.py", line 581, in run_server
    async with build_async_engine_client(args) as engine_client:
  File "/usr/lib/python3.10/contextlib.py", line 199, in __aenter__
    return await anext(self.gen)
  File "/home/jovyan/cp-sd-check/vllm/entrypoints/openai/api_server.py", line 107, in build_async_engine_client
    async with build_async_engine_client_from_engine_args(
  File "/usr/lib/python3.10/contextlib.py", line 199, in __aenter__
    return await anext(self.gen)
  File "/home/jovyan/cp-sd-check/vllm/entrypoints/openai/api_server.py", line 188, in build_async_engine_client_from_engine_args
    engine_config = engine_args.create_engine_config()
  File "/home/jovyan/cp-sd-check/vllm/engine/arg_utils.py", line 1071, in create_engine_config
    speculative_config = SpeculativeConfig.maybe_create_spec_config(
  File "/home/jovyan/cp-sd-check/vllm/config.py", line 1393, in maybe_create_spec_config
    raise ValueError(
ValueError: Chunked prefill and speculative decoding can be enabled simultaneously only for draft models with tensor parallel size 1.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[sroy745]

@njhill / @LiuXiaoxuanPKU this pr is ready for review. Could you ptal when you get a chance?

cc: @NickLucche

[LiuXiaoxuanPKU]

LGTM. Thanks!