diff --git a/.buildkite/performance-benchmarks/tests/latency-tests-arm64-cpu.json b/.buildkite/performance-benchmarks/tests/latency-tests-arm64-cpu.json index 98811049d3..d0693a19dc 100644 --- a/.buildkite/performance-benchmarks/tests/latency-tests-arm64-cpu.json +++ b/.buildkite/performance-benchmarks/tests/latency-tests-arm64-cpu.json @@ -3,7 +3,6 @@ "test_name": "latency_llama8B_tp1", "environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_KVCACHE_SPACE": 40 }, "parameters": { diff --git a/.buildkite/performance-benchmarks/tests/latency-tests-cpu.json b/.buildkite/performance-benchmarks/tests/latency-tests-cpu.json index 5f048df5f6..07e58346fc 100644 --- a/.buildkite/performance-benchmarks/tests/latency-tests-cpu.json +++ b/.buildkite/performance-benchmarks/tests/latency-tests-cpu.json @@ -3,7 +3,6 @@ "test_name": "latency_llama8B_tp2", "environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_SGL_KERNEL": 1, "VLLM_CPU_KVCACHE_SPACE": 40 }, diff --git a/.buildkite/performance-benchmarks/tests/serving-tests-arm64-cpu.json b/.buildkite/performance-benchmarks/tests/serving-tests-arm64-cpu.json index 75b80b2c21..d68837fa96 100644 --- a/.buildkite/performance-benchmarks/tests/serving-tests-arm64-cpu.json +++ b/.buildkite/performance-benchmarks/tests/serving-tests-arm64-cpu.json @@ -14,7 +14,6 @@ ], "server_environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_SGL_KERNEL": 1, "VLLM_CPU_KVCACHE_SPACE": 40 }, diff --git a/.buildkite/performance-benchmarks/tests/serving-tests-cpu-asr.json b/.buildkite/performance-benchmarks/tests/serving-tests-cpu-asr.json index 16e8b0600a..756f0f77f2 100644 --- a/.buildkite/performance-benchmarks/tests/serving-tests-cpu-asr.json +++ b/.buildkite/performance-benchmarks/tests/serving-tests-cpu-asr.json @@ -4,9 +4,6 @@ "inf" ], "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200], - "server_environment_variables": { - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120 - }, "server_parameters": { "dtype": "bfloat16", "model": "openai/whisper-large-v3-turbo" diff --git a/.buildkite/performance-benchmarks/tests/serving-tests-cpu-embed.json b/.buildkite/performance-benchmarks/tests/serving-tests-cpu-embed.json index c62f244fc7..1be4c653f1 100644 --- a/.buildkite/performance-benchmarks/tests/serving-tests-cpu-embed.json +++ b/.buildkite/performance-benchmarks/tests/serving-tests-cpu-embed.json @@ -10,7 +10,6 @@ ], "server_environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_SGL_KERNEL": 1, "VLLM_CPU_KVCACHE_SPACE": 40 }, diff --git a/.buildkite/performance-benchmarks/tests/serving-tests-cpu-text.json b/.buildkite/performance-benchmarks/tests/serving-tests-cpu-text.json index 9aa76c1108..acc3669dff 100644 --- a/.buildkite/performance-benchmarks/tests/serving-tests-cpu-text.json +++ b/.buildkite/performance-benchmarks/tests/serving-tests-cpu-text.json @@ -6,7 +6,6 @@ "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200], "server_environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_SGL_KERNEL": 1, "VLLM_CPU_KVCACHE_SPACE": 40 }, diff --git a/.buildkite/performance-benchmarks/tests/serving-tests-cpu.json b/.buildkite/performance-benchmarks/tests/serving-tests-cpu.json index 0b7e749996..f1b2f4747a 100644 --- a/.buildkite/performance-benchmarks/tests/serving-tests-cpu.json +++ b/.buildkite/performance-benchmarks/tests/serving-tests-cpu.json @@ -6,7 +6,6 @@ "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200], "server_environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_SGL_KERNEL": 1, "VLLM_CPU_KVCACHE_SPACE": 40 }, diff --git a/.buildkite/performance-benchmarks/tests/throughput-tests-arm64-cpu.json b/.buildkite/performance-benchmarks/tests/throughput-tests-arm64-cpu.json index 3863cccf43..3bc7dde979 100644 --- a/.buildkite/performance-benchmarks/tests/throughput-tests-arm64-cpu.json +++ b/.buildkite/performance-benchmarks/tests/throughput-tests-arm64-cpu.json @@ -3,7 +3,6 @@ "test_name": "throughput_llama8B_tp1", "environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_KVCACHE_SPACE": 40 }, "parameters": { diff --git a/.buildkite/performance-benchmarks/tests/throughput-tests-cpu.json b/.buildkite/performance-benchmarks/tests/throughput-tests-cpu.json index d3f16eff11..089be00eb4 100644 --- a/.buildkite/performance-benchmarks/tests/throughput-tests-cpu.json +++ b/.buildkite/performance-benchmarks/tests/throughput-tests-cpu.json @@ -3,7 +3,6 @@ "test_name": "throughput_llama8B_tp2", "environment_variables": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1, - "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120, "VLLM_CPU_SGL_KERNEL": 1, "VLLM_CPU_KVCACHE_SPACE": 40 }, diff --git a/tests/entrypoints/multimodal/openai/chat_completion/test_video.py b/tests/entrypoints/multimodal/openai/chat_completion/test_video.py index a6ca410e07..2431dd5339 100644 --- a/tests/entrypoints/multimodal/openai/chat_completion/test_video.py +++ b/tests/entrypoints/multimodal/openai/chat_completion/test_video.py @@ -40,13 +40,11 @@ def server(): "--trust-request-mm-kwargs", ] - # ROCm: Increase timeouts to handle potential network delays and slower - # video processing when downloading multiple videos from external sources + # ROCm: Allow more time to download videos from external sources. env_overrides = {} if current_platform.is_rocm(): env_overrides = { "VLLM_VIDEO_FETCH_TIMEOUT": "120", - "VLLM_ENGINE_ITERATION_TIMEOUT_S": "300", } with RemoteOpenAIServer(MODEL_NAME, args, env_dict=env_overrides) as remote_server: diff --git a/tests/entrypoints/multimodal/openai/chat_completion/test_vision.py b/tests/entrypoints/multimodal/openai/chat_completion/test_vision.py index 56a7cf6a13..6438d50a78 100644 --- a/tests/entrypoints/multimodal/openai/chat_completion/test_vision.py +++ b/tests/entrypoints/multimodal/openai/chat_completion/test_vision.py @@ -88,13 +88,11 @@ def server(): *ROCM_EXTRA_ARGS, ] - # ROCm: Increase timeouts to handle potential network delays and slower - # video processing when downloading multiple videos from external sources + # ROCm: Allow more time to download videos from external sources. env_overrides = { **( { "VLLM_VIDEO_FETCH_TIMEOUT": "120", - "VLLM_ENGINE_ITERATION_TIMEOUT_S": "300", } if current_platform.is_rocm() else {} diff --git a/tests/entrypoints/speech_to_text/realtime/test_realtime_validation.py b/tests/entrypoints/speech_to_text/realtime/test_realtime_validation.py index 7fb7a5522d..c74246b2f8 100644 --- a/tests/entrypoints/speech_to_text/realtime/test_realtime_validation.py +++ b/tests/entrypoints/speech_to_text/realtime/test_realtime_validation.py @@ -14,12 +14,6 @@ from tests.utils import ROCM_EXTRA_ARGS, RemoteOpenAIServer from vllm.assets.audio import AudioAsset from vllm.multimodal.media.audio import load_audio -# Increase engine iteration timeout for ROCm where first-use JIT compilation -# can exceed the default 60s, causing a silent deadlock in feed_tokens. -REALTIME_ENV_OVERRIDES = { - "VLLM_ENGINE_ITERATION_TIMEOUT_S": "600", -} - MISTRAL_FORMAT_ARGS = [ "--tokenizer_mode", "mistral", @@ -77,9 +71,7 @@ async def test_multi_chunk_streaming(model_name, mary_had_lamb_audio_chunks): if model_name.startswith("mistralai"): server_args += MISTRAL_FORMAT_ARGS - with RemoteOpenAIServer( - model_name, server_args, env_dict=REALTIME_ENV_OVERRIDES - ) as remote_server: + with RemoteOpenAIServer(model_name, server_args) as remote_server: ws_url = _get_websocket_url(remote_server) async with websockets.connect(ws_url) as ws: # Receive session.created @@ -183,9 +175,7 @@ async def test_empty_commit_does_not_crash_engine( if model_name.startswith("mistralai"): server_args += MISTRAL_FORMAT_ARGS - with RemoteOpenAIServer( - model_name, server_args, env_dict=REALTIME_ENV_OVERRIDES - ) as remote_server: + with RemoteOpenAIServer(model_name, server_args) as remote_server: ws_url = _get_websocket_url(remote_server) # --- First connection: empty commit (no audio appended) ---------- @@ -272,9 +262,7 @@ async def test_session_update_invalid_model_returns_error(model_name): if model_name.startswith("mistralai"): server_args += MISTRAL_FORMAT_ARGS - with RemoteOpenAIServer( - model_name, server_args, env_dict=REALTIME_ENV_OVERRIDES - ) as remote_server: + with RemoteOpenAIServer(model_name, server_args) as remote_server: ws_url = _get_websocket_url(remote_server) async with websockets.connect(ws_url) as ws: event = await receive_event(ws, timeout=30.0) @@ -301,9 +289,7 @@ async def test_commit_without_session_update_returns_error(model_name): if model_name.startswith("mistralai"): server_args += MISTRAL_FORMAT_ARGS - with RemoteOpenAIServer( - model_name, server_args, env_dict=REALTIME_ENV_OVERRIDES - ) as remote_server: + with RemoteOpenAIServer(model_name, server_args) as remote_server: ws_url = _get_websocket_url(remote_server) async with websockets.connect(ws_url) as ws: event = await receive_event(ws, timeout=30.0) diff --git a/vllm/config/ec_transfer.py b/vllm/config/ec_transfer.py index 5c7f99b4aa..5c2b8e589c 100644 --- a/vllm/config/ec_transfer.py +++ b/vllm/config/ec_transfer.py @@ -42,15 +42,6 @@ class ECTransferConfig: """Whether this vLLM instance produces, consumes EC cache, or both. Choices are 'ec_producer', 'ec_consumer', 'ec_both'.""" - ec_rank: int | None = None - """The rank of this vLLM instance in the EC cache transfer. Typical value: - 0 for encoder, 1 for pd instance. - Currently only 1P1D is supported.""" - - ec_parallel_size: int = 1 - """The number of parallel instances for EC cache transfer. For - PyNcclConnector, this should be 2.""" - ec_ip: str = "127.0.0.1" """The EC connector ip, used to build distributed connection.""" diff --git a/vllm/config/kv_transfer.py b/vllm/config/kv_transfer.py index dfc9e02f3d..59eb4c80be 100644 --- a/vllm/config/kv_transfer.py +++ b/vllm/config/kv_transfer.py @@ -68,25 +68,10 @@ class KVTransferConfig: """The device used by kv connector to buffer the KV cache. Choices are 'cuda', 'cpu' and 'xpu'.""" - kv_buffer_size: float = 1e9 - """The buffer size for TorchDistributedConnector. Measured in number of - bytes. Recommended value: 1e9 (about 1GB).""" - kv_role: KVRole | None = None """Whether this vLLM instance produces, consumes KV cache, or both. Choices are 'kv_producer', 'kv_consumer', and 'kv_both'.""" - kv_rank: int | None = None - """The rank of this vLLM instance in the KV cache transfer. Typical value: - 0 for prefill instance, 1 for decode instance. - Currently only 1P1D is supported.""" - - kv_parallel_size: int = 1 - """The number of parallel instances for KV cache transfer.""" - - kv_ip: str = "127.0.0.1" - """The KV connector ip, used to build distributed connection.""" - kv_port: int = 14579 """The KV connector port, used to build distributed connection.""" diff --git a/vllm/config/parallel.py b/vllm/config/parallel.py index 53a28fa691..7adc48254b 100644 --- a/vllm/config/parallel.py +++ b/vllm/config/parallel.py @@ -277,9 +277,6 @@ class ParallelConfig: worker_cls: str = "auto" """The full name of the worker class to use. If "auto", the worker class will be determined based on the platform.""" - sd_worker_cls: str = "auto" - """The full name of the worker class to use for speculative decoding. - If "auto", the worker class will be determined based on the platform.""" worker_extension_cls: str = "" """The full name of the worker extension class to use. The worker extension class is dynamically inherited by the worker class. This is used to inject @@ -894,7 +891,6 @@ class ParallelConfig: "placement_group", "distributed_executor_backend", "worker_cls", - "sd_worker_cls", "worker_extension_cls", "_api_process_count", "_api_process_rank", diff --git a/vllm/envs.py b/vllm/envs.py index e63ed5cc02..ac8ba21682 100755 --- a/vllm/envs.py +++ b/vllm/envs.py @@ -23,7 +23,6 @@ if TYPE_CHECKING: VLLM_ROCM_SLEEP_MEM_CHUNK_SIZE: int = 256 LOCAL_RANK: int = 0 CUDA_VISIBLE_DEVICES: str | None = None - VLLM_ENGINE_ITERATION_TIMEOUT_S: int = 60 VLLM_ENGINE_READY_TIMEOUT_S: int = 600 VLLM_CHAT_TEMPLATE_RENDER_TIMEOUT: float = 30.0 VLLM_API_KEY: str | None = None @@ -166,9 +165,6 @@ if TYPE_CHECKING: VLLM_DISABLE_COMPILE_CACHE: bool = False VLLM_REPLICATE_EMBED: bool = False VLLM_USE_LAYERNAME: bool = True - Q_SCALE_CONSTANT: int = 200 - K_SCALE_CONSTANT: int = 200 - V_SCALE_CONSTANT: int = 100 VLLM_USE_RUST_FRONTEND: bool = False VLLM_USE_RUST_BENCH: bool = False VLLM_RUST_FRONTEND_PATH: str | None = "auto" @@ -804,10 +800,6 @@ environment_variables: dict[str, Callable[[], Any]] = { "LOCAL_RANK": lambda: int(os.environ.get("LOCAL_RANK", "0")), # used to control the visible devices in the distributed setting "CUDA_VISIBLE_DEVICES": lambda: os.environ.get("CUDA_VISIBLE_DEVICES", None), - # timeout for each iteration in the engine - "VLLM_ENGINE_ITERATION_TIMEOUT_S": lambda: int( - os.environ.get("VLLM_ENGINE_ITERATION_TIMEOUT_S", "60") - ), # Timeout in seconds for waiting for engine cores to become ready # during startup. Default is 600 seconds (10 minutes). "VLLM_ENGINE_READY_TIMEOUT_S": lambda: int( @@ -2368,7 +2360,6 @@ def compile_factors() -> dict[str, object]: "VLLM_TUNED_CONFIG_FOLDER", "VLLM_FLASHINFER_AUTOTUNE_CACHE_DIR", "VLLM_FLASHINFER_AUTOTUNE_SKIP_OPS", - "VLLM_ENGINE_ITERATION_TIMEOUT_S", "VLLM_HTTP_TIMEOUT_KEEP_ALIVE", "VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS", "VLLM_WORKER_SHUTDOWN_TIMEOUT_SECONDS",