diff --git a/vllm/v1/engine/core_client.py b/vllm/v1/engine/core_client.py index e19b31396b..7e1f1cf418 100644 --- a/vllm/v1/engine/core_client.py +++ b/vllm/v1/engine/core_client.py @@ -609,8 +609,13 @@ class MPClient(EngineCoreClient): timeout=VLLM_ENGINE_READY_TIMEOUT_S * 1000 # convert to ms ): raise TimeoutError( - "Timed out waiting for engines to send " - "initial message on input socket." + f"Timed out waiting for engine core processes to " + f"start. This is often caused by slow weight loading " + f"for large models. Waited " + f"{VLLM_ENGINE_READY_TIMEOUT_S}s (configured by " + f"VLLM_ENGINE_READY_TIMEOUT_S). To increase the " + f"timeout, set the environment variable: " + f"VLLM_ENGINE_READY_TIMEOUT_S=" ) identity, _ = sync_input_socket.recv_multipart() identities.remove(identity) @@ -1586,8 +1591,12 @@ class DPLBAsyncMPClient(DPAsyncMPClient): timeout=VLLM_ENGINE_READY_TIMEOUT_S * 1000 # convert to ms ): raise TimeoutError( - "Timed out waiting for new engines to send initial " - "message on input socket." + f"Timed out waiting for new engine core processes to " + f"start. Waited " + f"{VLLM_ENGINE_READY_TIMEOUT_S}s (configured by " + f"VLLM_ENGINE_READY_TIMEOUT_S). To increase the " + f"timeout, set the environment variable: " + f"VLLM_ENGINE_READY_TIMEOUT_S=" ) identity, _ = sync_input_socket.recv_multipart() new_engine_identities.discard(identity)