diff --git a/vllm/v1/engine/__init__.py b/vllm/v1/engine/__init__.py index da26daffdd..c97adfe8f4 100644 --- a/vllm/v1/engine/__init__.py +++ b/vllm/v1/engine/__init__.py @@ -72,9 +72,9 @@ class EngineCoreReadyResponse: values (e.g. max_model_len after KV cache auto-fitting). """ + max_model_len: int num_gpu_blocks: int dp_stats_address: str | None - max_model_len: int | None = None class EngineCoreRequest( diff --git a/vllm/v1/engine/core_client.py b/vllm/v1/engine/core_client.py index 5d170b2171..2f2c15d246 100644 --- a/vllm/v1/engine/core_client.py +++ b/vllm/v1/engine/core_client.py @@ -671,11 +671,9 @@ class MPClient(EngineCoreClient): return vllm_config = self.vllm_config response = msgspec.msgpack.decode(payload, type=EngineCoreReadyResponse) - if response.max_model_len is not None: - vllm_config.model_config.max_model_len = min( - vllm_config.model_config.max_model_len, - response.max_model_len, - ) + vllm_config.model_config.max_model_len = min( + vllm_config.model_config.max_model_len, response.max_model_len + ) # Setup KV cache config with initialization state from # engine core process. Sum values from all engines in DP case.