diff --git a/tests/test_config.py b/tests/test_config.py index e97e2cebaa..7a6ea1d4c4 100644 --- a/tests/test_config.py +++ b/tests/test_config.py @@ -1392,16 +1392,8 @@ def test_engram_model_support( assert resolved.engram_config.cpu_offload is True -@pytest.mark.parametrize( - ("value", "expected"), [(None, True), ("0", False), ("1", True)] -) -def test_engram_cpu_offload_environment_default(monkeypatch, value, expected): - monkeypatch.delenv("VLLM_PLE_CPU_OFFLOAD", raising=False) - if value is not None: - monkeypatch.setenv("VLLM_PLE_CPU_OFFLOAD", value) - assert EngramConfig().cpu_offload is expected - assert EngramConfig(cpu_offload=False).cpu_offload is False - assert EngramConfig(cpu_offload=True).cpu_offload is True +def test_engram_cpu_offload_default(): + assert EngramConfig().cpu_offload is True def test_engram_config_defaults_to_none(): diff --git a/vllm/config/engram.py b/vllm/config/engram.py index c9f05c9382..0f5ec31b33 100644 --- a/vllm/config/engram.py +++ b/vllm/config/engram.py @@ -3,10 +3,9 @@ from typing import TYPE_CHECKING -from pydantic import Field, model_validator +from pydantic import model_validator from typing_extensions import Self -import vllm.envs as envs from vllm.config.utils import config, get_hash_factors, hash_factors if TYPE_CHECKING: @@ -22,10 +21,6 @@ _NGRAM_LAYER_FIELDS = { } -def _default_cpu_offload() -> bool: - return envs.VLLM_PLE_CPU_OFFLOAD - - def model_has_engram_layers(model_config: "ModelConfig | None") -> bool: """Whether the model carries n-gram embedding layers.""" if model_config is None: @@ -40,10 +35,8 @@ def model_has_engram_layers(model_config: "ModelConfig | None") -> bool: class EngramConfig: """Configuration for Engram embedding storage and sharding.""" - cpu_offload: bool = Field(default_factory=_default_cpu_offload) - """Store embedding weights in pinned CPU memory for UVA lookup. - Defaults to VLLM_PLE_CPU_OFFLOAD, which is enabled by default. An explicit - value takes precedence over the environment variable.""" + cpu_offload: bool = True + """Store embedding weights in pinned CPU memory for UVA lookup.""" embedding_across_dp: bool = False """Shard embeddings across TP and all DP ranks when enabled. diff --git a/vllm/envs.py b/vllm/envs.py index 259c08e7e7..fa98db1cf1 100755 --- a/vllm/envs.py +++ b/vllm/envs.py @@ -156,7 +156,6 @@ if TYPE_CHECKING: VLLM_ROCM_SHUFFLE_KV_CACHE_LAYOUT: bool = False VLLM_ENABLE_V1_MULTIPROCESSING: bool = True VLLM_LOG_BATCHSIZE_INTERVAL: float = -1 - VLLM_PLE_CPU_OFFLOAD: bool = True VLLM_DISABLE_COMPILE_CACHE: bool = False VLLM_REPLICATE_EMBED: bool = False VLLM_USE_LAYERNAME: bool = True @@ -2066,8 +2065,6 @@ environment_variables: dict[str, Callable[[], Any]] = { "VLLM_LOG_MODEL_INSPECTION": lambda: bool( int(os.getenv("VLLM_LOG_MODEL_INSPECTION", "0")) ), - # Store n-gram embedding tables in pinned CPU memory for UVA lookup. - "VLLM_PLE_CPU_OFFLOAD": lambda: bool(int(os.getenv("VLLM_PLE_CPU_OFFLOAD", "1"))), # Debug logging for --enable-mfu-metrics "VLLM_DEBUG_MFU_METRICS": lambda: bool( int(os.getenv("VLLM_DEBUG_MFU_METRICS", "0"))