@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 8bab4959be
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 16, 23:15 UTC

[Misc] Remove VLLM_BUILD_WITH_NEURON env variable (#5389)

Woosuk Kwon committed Jun 11, 2024 at 00:37 UTC 8bab4959bea640f8f81ca59eb06b1f056ac23111
3 files changed +2 -7
Dockerfile.neuron
+1 -1
@@ -28,7 +28,7 @@ COPY ./requirements-neuron.txt /app/vllm/requirements-neuron.txt
28 RUN cd /app/vllm \
29 && python3 -m pip install -U -r requirements-neuron.txt
30
31 -ENV VLLM_BUILD_WITH_NEURON 1
31 +ENV VLLM_TARGET_DEVICE neuron
32 RUN cd /app/vllm \
33 && pip install -e . \
34 && cd ..
setup.py
+1 -1
@@ -222,7 +222,7 @@ def _is_neuron() -> bool:
222 subprocess.run(["neuron-ls"], capture_output=True, check=True)
223 except (FileNotFoundError, PermissionError, subprocess.CalledProcessError):
224 torch_neuronx_installed = False
225 - return torch_neuronx_installed or envs.VLLM_BUILD_WITH_NEURON
225 + return torch_neuronx_installed or VLLM_TARGET_DEVICE == "neuron"
226
227
228 def _is_cpu() -> bool:
vllm/envs.py
-5
@@ -33,7 +33,6 @@ if TYPE_CHECKING:
33 VLLM_TARGET_DEVICE: str = "cuda"
34 MAX_JOBS: Optional[str] = None
35 NVCC_THREADS: Optional[str] = None
36 - VLLM_BUILD_WITH_NEURON: bool = False
36 VLLM_USE_PRECOMPILED: bool = False
37 VLLM_INSTALL_PUNICA_KERNELS: bool = False
38 CMAKE_BUILD_TYPE: Optional[str] = None
@@ -63,10 +62,6 @@ environment_variables: Dict[str, Callable[[], Any]] = {
62 "NVCC_THREADS":
63 lambda: os.getenv("NVCC_THREADS", None),
64
66 - # If set, vllm will build with Neuron support
67 - "VLLM_BUILD_WITH_NEURON":
68 - lambda: bool(os.environ.get("VLLM_BUILD_WITH_NEURON", False)),
69 -
65 # If set, vllm will use precompiled binaries (*.so)
66 "VLLM_USE_PRECOMPILED":
67 lambda: bool(os.environ.get("VLLM_USE_PRECOMPILED")),