vllm - 💡(How to fix) Fix [Bug]: max-num-partial-prefills failes on V1 engine start [1 participants]

vllm serve /tmp/model-poc --host 127.0.0.1 --port 8000 --gpu-memory-utilization 0.90 --max-model-len 1000 --dtype float16 --kv-cache-dtype auto --tensor-parallel-size 1 --pipeline-parallel-size 1 --max-num-batched-tokens 4096 --max-num-seqs 128 --max-num-partial-prefills 4 --max-long-partial-prefills 1 --long-prefill-token-threshold 256 --performance-mode interactivity

2026-04-13 21:12:29,527 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [utils.py:299] 2026-04-13 21:12:29,527 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [utils.py:299] █ █ █▄ ▄█ 2026-04-13 21:12:29,527 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [utils.py:299] ▄▄ ▄█ █ █ █ ▀▄▀ █ version 0.19.0 2026-04-13 21:12:29,527 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [utils.py:299] █▄█▀ █ █ █ █ model /tmp/model-poc 2026-04-13 21:12:29,527 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [utils.py:299] ▀▀ ▀▀▀▀▀ ▀▀▀▀▀ ▀ ▀ 2026-04-13 21:12:29,527 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [utils.py:299] 2026-04-13 21:12:29,531 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [utils.py:233] non-default args: {'model_tag': '/tmp/model-load-testing-poc/model-poc', 'host': '127.0.0.1', 'model': '/tmp/model-poc', 'dtype': 'float16', 'max_model_len': 1000, 'max_num_batched_tokens': 4096, 'max_num_seqs': 128, 'max_num_partial_prefills': 4, 'long_prefill_token_threshold': 256, 'performance_mode': 'interactivity'} 2026-04-13 21:12:29,540 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [model.py:549] Resolved architecture: Qwen3ForCausalLM 2026-04-13 21:12:29,540 INFO [server] (APIServer pid=6907) WARNING 04-13 21:12:29 [model.py:2016] Casting torch.bfloat16 to torch.float16. 2026-04-13 21:12:29,540 INFO [server] (APIServer pid=6907) INFO 04-13 21:12:29 [model.py:1678] Using max model len 1000 2026-04-13 21:12:29,541 INFO [server] (APIServer pid=6907) Traceback (most recent call last): 2026-04-13 21:12:29,541 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/bin/vllm", line 6, in <module> 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) sys.exit(main()) 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) ^^^^^^ 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/entrypoints/cli/main.py", line 75, in main 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) args.dispatch_function(args) 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/entrypoints/cli/serve.py", line 122, in cmd 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) uvloop.run(run_server(args)) 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/uvloop/init.py", line 92, in run 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) return runner.run(wrapper()) 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) ^^^^^^^^^^^^^^^^^^^^^ 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/asyncio/runners.py", line 118, in run 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) return self._loop.run_until_complete(task) 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/uvloop/init.py", line 48, in wrapper 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) return await main 2026-04-13 21:12:29,542 INFO [server] (APIServer pid=6907) ^^^^^^^^^^ 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/entrypoints/openai/api_server.py", line 670, in run_server 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) await run_server_worker(listen_address, sock, args, **uvicorn_kwargs) 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/entrypoints/openai/api_server.py", line 684, in run_server_worker 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) async with build_async_engine_client( 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/contextlib.py", line 210, in aenter 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) return await anext(self.gen) 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) ^^^^^^^^^^^^^^^^^^^^^ 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/entrypoints/openai/api_server.py", line 100, in build_async_engine_client 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) async with build_async_engine_client_from_engine_args( 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/contextlib.py", line 210, in aenter 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) return await anext(self.gen) 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) ^^^^^^^^^^^^^^^^^^^^^ 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/entrypoints/openai/api_server.py", line 124, in build_async_engine_client_from_engine_args 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) vllm_config = engine_args.create_engine_config(usage_context=usage_context) 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2026-04-13 21:12:29,543 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/engine/arg_utils.py", line 1554, in create_engine_config 2026-04-13 21:12:29,544 INFO [server] (APIServer pid=6907) self._check_feature_supported() 2026-04-13 21:12:29,544 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/engine/arg_utils.py", line 2004, in _check_feature_supported 2026-04-13 21:12:29,544 INFO [server] (APIServer pid=6907) _raise_unsupported_error(feature_name="Concurrent Partial Prefill") 2026-04-13 21:12:29,544 INFO [server] (APIServer pid=6907) File "/root/miniconda3/envs/test/lib/python3.11/site-packages/vllm/engine/arg_utils.py", line 2283, in _raise_unsupported_error 2026-04-13 21:12:29,544 INFO [server] (APIServer pid=6907) raise NotImplementedError(msg) 2026-04-13 21:12:29,544 INFO [server] (APIServer pid=6907) NotImplementedError: Concurrent Partial Prefill is not supported. We recommend to remove Concurrent Partial Prefill from your config.

TL;DR

The issue is likely due to the max-num-partial-prefills parameter being enabled, which is not supported in the current version, and removing it from the config should resolve the issue.

Guidance

• The error message NotImplementedError: Concurrent Partial Prefill is not supported suggests that the max-num-partial-prefills parameter is not supported in the current version.
• To fix the issue, try removing the --max-num-partial-prefills parameter from the command line arguments.
• Verify that the max-num-partial-prefills parameter is not set in any configuration files or environment variables.
• If the issue persists, check the documentation for any updates on supported parameters and configurations.

Example

No code snippet is provided as the issue is related to command line arguments and configuration.

Notes

The max-num-partial-prefills parameter is not supported in the current version, and removing it should resolve the issue. However, this may affect the performance or behavior of the application.

Recommendation

Apply workaround: Remove the --max-num-partial-prefills parameter from the command line arguments to resolve the issue.