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 3, 03:00 UTC
Disable build isolation to bypass CUDA related deps for vllm-tpu (#43038)
Signed-off-by: Ylang Tsou <ylangt@google.com> Co-authored-by: Ylang Tsou <ylangt@google.com> Co-authored-by: Michael Goin <mgoin64@gmail.com>
ylangtsou committed
May 22, 2026 at 06:00 UTC
0b59fc45dd475f96f6f46f2c3e699d7bc13b3b04
2 files changed
+11
-1
requirements/build/tpu.txt
new
+7
@@ -0,0 +1,7 @@
1
+--extra-index-url https://download.pytorch.org/whl/cpu
2
+cmake>=3.26.1
3
+ninja
4
+setuptools>=77.0.3,<81.0.0
5
+setuptools-scm>=8
6
+torch==2.11.0+cpu
7
+wheel
tools/vllm-tpu/build.sh
+4
-1
@@ -79,12 +79,15 @@ trap cleanup EXIT HUP INT QUIT PIPE TERM # Register cleanup function to run on s
79
80
echo "Updating pyproject.toml completed. Proceeding with build..."
81
82
+echo "Install dependencies for no-isolation build"
83
+pip install -r "$VLLM_DIR/requirements/build/tpu.txt"
84
+
85
echo "Building wheel for TPU..."
86
rm -rf dist/
87
mkdir -p dist/
88
89
# User confirmed to use 'python -m build' directly
87
-if ! VLLM_TARGET_DEVICE=tpu python -m build; then
90
+if ! VLLM_TARGET_DEVICE=tpu python -m build --no-isolation; then
91
echo "Error: Python build command failed. Check if 'python -m build' works and the 'build' module is installed."
92
exit 1
93
fi