Revert #28875 (#29159)

docker/Dockerfile

@@ -56,6 +56,7 @@ ARG UV_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL}
 
 # PyTorch provides its own indexes for standard and nightly builds
 ARG PYTORCH_CUDA_INDEX_BASE_URL=https://download.pytorch.org/whl
+ARG PYTORCH_CUDA_NIGHTLY_INDEX_BASE_URL=https://download.pytorch.org/whl/nightly
 
 # PIP supports multiple authentication schemes, including keyring
 # By parameterizing the PIP_KEYRING_PROVIDER variable and setting it to
@@ -97,6 +98,7 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
 ARG PIP_INDEX_URL UV_INDEX_URL
 ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL
 ARG PYTORCH_CUDA_INDEX_BASE_URL
+ARG PYTORCH_CUDA_NIGHTLY_INDEX_BASE_URL
 ARG PIP_KEYRING_PROVIDER UV_KEYRING_PROVIDER
 
 # Activate virtual environment and add uv to PATH
@@ -315,6 +317,7 @@ RUN echo 'tzdata tzdata/Areas select America' | debconf-set-selections \
 ARG PIP_INDEX_URL UV_INDEX_URL
 ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL
 ARG PYTORCH_CUDA_INDEX_BASE_URL
+ARG PYTORCH_CUDA_NIGHTLY_INDEX_BASE_URL
 ARG PIP_KEYRING_PROVIDER UV_KEYRING_PROVIDER
 
 # Install uv for faster pip installs
@@ -334,6 +337,20 @@ ENV UV_LINK_MODE=copy
 # or future versions of triton.
 RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/
 
+# arm64 (GH200) build follows the practice of "use existing pytorch" build,
+# we need to install torch and torchvision from the nightly builds first,
+# pytorch will not appear as a vLLM dependency in all of the following steps
+# after this step
+RUN --mount=type=cache,target=/root/.cache/uv \
+    if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
+        uv pip install --system \
+            --index-url ${PYTORCH_CUDA_NIGHTLY_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') \
+            "torch==2.8.0.dev20250318+cu128" "torchvision==0.22.0.dev20250319" ; \
+        uv pip install --system \
+            --index-url ${PYTORCH_CUDA_NIGHTLY_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') \
+            --pre pytorch_triton==3.3.0+gitab727c40 ; \
+    fi
+
 # Install vllm wheel first, so that torch etc will be installed.
 RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist \
     --mount=type=cache,target=/root/.cache/uv \

docs/deployment/docker.md

@@ -82,7 +82,8 @@ DOCKER_BUILDKIT=1 docker build . \
 
 ## Building for Arm64/aarch64
 
-A docker container can be built for aarch64 systems such as the Nvidia Grace-Hopper. At time of this writing, this should be considered **experimental**. Using the flag `--platform "linux/arm64"` will attempt to build for arm64.
+A docker container can be built for aarch64 systems such as the Nvidia Grace-Hopper. At time of this writing, this requires the use
+of PyTorch Nightly and should be considered **experimental**. Using the flag `--platform "linux/arm64"` will attempt to build for arm64.
 
 !!! note
     Multiple modules must be compiled, so this process can take a while. Recommend using `--build-arg max_jobs=` & `--build-arg nvcc_threads=`
@@ -93,6 +94,7 @@ A docker container can be built for aarch64 systems such as the Nvidia Grace-Hop
 
     ```bash
     # Example of building on Nvidia GH200 server. (Memory usage: ~15GB, Build time: ~1475s / ~25 min, Image size: 6.93GB)
+    python3 use_existing_torch.py
     DOCKER_BUILDKIT=1 docker build . \
     --file docker/Dockerfile \
     --target vllm-openai \
@@ -100,8 +102,7 @@ A docker container can be built for aarch64 systems such as the Nvidia Grace-Hop
     -t vllm/vllm-gh200-openai:latest \
     --build-arg max_jobs=66 \
     --build-arg nvcc_threads=2 \
-    --build-arg torch_cuda_arch_list="9.0 10.0+PTX" \
-    --build-arg RUN_WHEEL_CHECK=false
+    --build-arg torch_cuda_arch_list="9.0 10.0+PTX"
     ```
 
 !!! note

docs/getting_started/installation/gpu.cuda.inc.md

@@ -158,7 +158,10 @@ uv pip install -e .
 
 ##### Use an existing PyTorch installation
 
-There are scenarios where the PyTorch dependency cannot be easily installed with `uv`, for example, when building vLLM with non-default PyTorch builds (like nightly or a custom build).
+There are scenarios where the PyTorch dependency cannot be easily installed with `uv`, e.g.:
+
+- Building vLLM with PyTorch nightly or a custom PyTorch build.
+- Building vLLM with aarch64 and CUDA (GH200), where the PyTorch wheels are not available on PyPI. Currently, only the PyTorch nightly has wheels for aarch64 with CUDA. You can run `uv pip install --index-url https://download.pytorch.org/whl/nightly/cu128 torch torchvision torchaudio` to [install PyTorch nightly](https://pytorch.org/get-started/locally/) and then build vLLM on top of it.
 
 To build vLLM using an existing PyTorch installation: