diff --git a/docker/Dockerfile b/docker/Dockerfile
index 006481b23cb9..8bcd7f118f1e 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -150,8 +150,8 @@ ARG torch_cuda_arch_list='7.0 7.5 8.0 8.9 9.0 10.0 12.0'
 ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
 #################### BASE BUILD IMAGE ####################
 
-#################### WHEEL BUILD IMAGE ####################
-FROM base AS build
+#################### CSRC BUILD IMAGE ####################
+FROM base AS csrc-build
 ARG TARGETPLATFORM
 
 ARG PIP_INDEX_URL UV_INDEX_URL
@@ -172,10 +172,13 @@ RUN --mount=type=cache,target=/root/.cache/uv \
     uv pip install --python /opt/venv/bin/python3 -r requirements/build.txt \
     --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
 
-COPY . .
-ARG GIT_REPO_CHECK=0
-RUN --mount=type=bind,source=.git,target=.git \
-    if [ "$GIT_REPO_CHECK" != "0" ]; then bash tools/check_repo.sh ; fi
+WORKDIR /workspace
+
+COPY pyproject.toml setup.py CMakeLists.txt ./
+COPY cmake cmake/
+COPY csrc csrc/
+COPY vllm/envs.py vllm/envs.py
+COPY vllm/__init__.py vllm/__init__.py
 
 # max jobs used by Ninja to build extensions
 ARG max_jobs=2
@@ -195,9 +198,11 @@ ARG SCCACHE_S3_NO_CREDENTIALS=0
 ARG VLLM_USE_PRECOMPILED=""
 ARG VLLM_MAIN_CUDA_VERSION=""
 
+# Use dummy version for csrc-build wheel (only .so files are extracted, version doesn't matter)
+ENV SETUPTOOLS_SCM_PRETEND_VERSION="0.0.0+csrc.build"
+
 # if USE_SCCACHE is set, use sccache to speed up compilation
 RUN --mount=type=cache,target=/root/.cache/uv \
-    --mount=type=bind,source=.git,target=.git \
     if [ "$USE_SCCACHE" = "1" ]; then \
         echo "Installing sccache..." \
         && curl -L -o sccache.tar.gz ${SCCACHE_DOWNLOAD_URL} \
@@ -223,7 +228,6 @@ ENV VLLM_TARGET_DEVICE=${vllm_target_device}
 ENV CCACHE_DIR=/root/.cache/ccache
 RUN --mount=type=cache,target=/root/.cache/ccache \
     --mount=type=cache,target=/root/.cache/uv \
-    --mount=type=bind,source=.git,target=.git  \
     if [ "$USE_SCCACHE" != "1" ]; then \
         # Clean any existing CMake artifacts
         rm -rf .deps && \
@@ -232,6 +236,52 @@ RUN --mount=type=cache,target=/root/.cache/ccache \
         export VLLM_DOCKER_BUILD_CONTEXT=1 && \
         python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38; \
     fi
+#################### CSRC BUILD IMAGE ####################
+
+#################### WHEEL BUILD IMAGE ####################
+FROM base AS build
+ARG TARGETPLATFORM
+
+ARG PIP_INDEX_URL UV_INDEX_URL
+ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL
+ARG PYTORCH_CUDA_INDEX_BASE_URL
+
+# install build dependencies
+COPY requirements/build.txt requirements/build.txt
+
+# This timeout (in seconds) is necessary when installing some dependencies via uv since it's likely to time out
+# Reference: https://github.com/astral-sh/uv/pull/1694
+ENV UV_HTTP_TIMEOUT=500
+ENV UV_INDEX_STRATEGY="unsafe-best-match"
+# Use copy mode to avoid hardlink failures with Docker cache mounts
+ENV UV_LINK_MODE=copy
+
+RUN --mount=type=cache,target=/root/.cache/uv \
+    uv pip install --python /opt/venv/bin/python3 -r requirements/build.txt \
+    --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
+
+WORKDIR /workspace
+
+COPY --from=csrc-build /workspace/dist /precompiled-wheels
+
+COPY . .
+
+ARG GIT_REPO_CHECK=0
+RUN --mount=type=bind,source=.git,target=.git \
+    if [ "$GIT_REPO_CHECK" != "0" ]; then bash tools/check_repo.sh ; fi
+
+ARG vllm_target_device="cuda"
+ENV VLLM_TARGET_DEVICE=${vllm_target_device}
+
+# Skip adding +precompiled suffix to version (preserves git-derived version)
+ENV VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX=1
+
+RUN --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,source=.git,target=.git \
+    if [ "${vllm_target_device}" = "cuda" ]; then \
+        export VLLM_PRECOMPILED_WHEEL_LOCATION=$(ls /precompiled-wheels/*.whl); \
+    fi && \
+    python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38
 
 # Install DeepGEMM from source
 ARG DEEPGEMM_GIT_REF
diff --git a/docs/assets/contributing/dockerfile-stages-dependency.png b/docs/assets/contributing/dockerfile-stages-dependency.png
index b327eb2151f5..7420ca4d8944 100644
Binary files a/docs/assets/contributing/dockerfile-stages-dependency.png and b/docs/assets/contributing/dockerfile-stages-dependency.png differ
diff --git a/setup.py b/setup.py
index 5b7d12bb373e..8b2b4f7e5def 100644
--- a/setup.py
+++ b/setup.py
@@ -461,14 +461,22 @@ class precompiled_wheel_utils:
                     "vllm/cumem_allocator.abi3.so",
                 ]
 
-                compiled_regex = re.compile(
+                flash_attn_regex = re.compile(
                     r"vllm/vllm_flash_attn/(?:[^/.][^/]*/)*(?!\.)[^/]*\.py"
                 )
+                triton_kernels_regex = re.compile(
+                    r"vllm/third_party/triton_kernels/(?:[^/.][^/]*/)*(?!\.)[^/]*\.py"
+                )
                 file_members = list(
                     filter(lambda x: x.filename in files_to_copy, wheel.filelist)
                 )
                 file_members += list(
-                    filter(lambda x: compiled_regex.match(x.filename), wheel.filelist)
+                    filter(lambda x: flash_attn_regex.match(x.filename), wheel.filelist)
+                )
+                file_members += list(
+                    filter(
+                        lambda x: triton_kernels_regex.match(x.filename), wheel.filelist
+                    )
                 )
 
                 for file in file_members:
@@ -648,7 +656,7 @@ def get_vllm_version() -> str:
         if envs.VLLM_TARGET_DEVICE == "empty":
             version += f"{sep}empty"
     elif _is_cuda():
-        if envs.VLLM_USE_PRECOMPILED:
+        if envs.VLLM_USE_PRECOMPILED and not envs.VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX:
             version += f"{sep}precompiled"
         else:
             cuda_version = str(get_nvcc_cuda_version())
diff --git a/vllm/envs.py b/vllm/envs.py
index 8b954fa14f28..4b594e54f721 100755
--- a/vllm/envs.py
+++ b/vllm/envs.py
@@ -78,6 +78,7 @@ if TYPE_CHECKING:
     MAX_JOBS: str | None = None
     NVCC_THREADS: str | None = None
     VLLM_USE_PRECOMPILED: bool = False
+    VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX: bool = False
     VLLM_DOCKER_BUILD_CONTEXT: bool = False
     VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL: bool = False
     VLLM_KEEP_ALIVE_ON_ENGINE_DEATH: bool = False
@@ -462,6 +463,10 @@ environment_variables: dict[str, Callable[[], Any]] = {
     .lower()
     in ("1", "true")
     or bool(os.environ.get("VLLM_PRECOMPILED_WHEEL_LOCATION")),
+    # If set, skip adding +precompiled suffix to version string
+    "VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX": lambda: bool(
+        int(os.environ.get("VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX", "0"))
+    ),
     # Used to mark that setup.py is running in a Docker build context,
     # in order to force the use of precompiled binaries.
     "VLLM_DOCKER_BUILD_CONTEXT": lambda: os.environ.get("VLLM_DOCKER_BUILD_CONTEXT", "")