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 2, 13:30 UTC
[ROCm] Bump ROCm to 7.2.3 (#43136)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
May 28, 2026 at 11:42 UTC
1b5437cec8101677028085fc512b078839051612
2 files changed
+3
-26
.buildkite/release-pipeline.yaml
+1
-1
@@ -737,7 +737,7 @@ steps:
737
- "bash tools/vllm-rocm/generate-rocm-wheels-root-index.sh"
738
env:
739
S3_BUCKET: "vllm-wheels"
740
- VARIANT: "rocm722"
740
+ VARIANT: "rocm723"
741
742
# ROCm Job 6: Build ROCm Release Docker Image
743
- label: ":docker: Build release image - x86_64 - ROCm"
docker/Dockerfile.rocm_base
+2
-25
@@ -1,4 +1,4 @@
1
-ARG BASE_IMAGE=rocm/dev-ubuntu-22.04:7.2.2-complete
1
+ARG BASE_IMAGE=rocm/dev-ubuntu-22.04:7.2.3-complete
2
ARG TRITON_BRANCH="ba5c1517"
3
ARG TRITON_REPO="https://github.com/ROCm/triton.git"
4
ARG PYTORCH_BRANCH="8514f051" # release/2.10 as of 3/17
@@ -104,28 +104,6 @@ ENV SCCACHE_REGION=${USE_SCCACHE:+${SCCACHE_REGION_NAME}}
104
ENV SCCACHE_S3_NO_CREDENTIALS=${USE_SCCACHE:+${SCCACHE_S3_NO_CREDENTIALS}}
105
ENV SCCACHE_IDLE_TIMEOUT=${USE_SCCACHE:+0}
106
107
-# torch profiler hotfix for 7.2.2: rebuild CLR with https://github.com/ROCm/rocm-systems/pull/5062
108
-# will be removed once we move to ROCm 7.2.3
109
-RUN apt-get update && apt-get install -y rocm-llvm-dev
110
-RUN pip install CppHeaderParser
111
-RUN git clone --no-checkout --filter=blob:none https://github.com/ROCm/rocm-systems /tmp/rocm-systems \
112
- && cd /tmp/rocm-systems \
113
- && git sparse-checkout init --cone \
114
- && git sparse-checkout set projects/hip projects/clr \
115
- && git checkout 35e8c7bf8911862e5389509800e65fdf125412b3 \
116
- && export CLR_DIR=/tmp/rocm-systems/projects/clr \
117
- && export HIP_DIR=/tmp/rocm-systems/projects/hip \
118
- && mkdir -p $CLR_DIR/build && cd $CLR_DIR/build \
119
- && cmake \
120
- -DHIP_COMMON_DIR=$HIP_DIR \
121
- -DCMAKE_PREFIX_PATH="/opt/rocm/" \
122
- -DCLR_BUILD_HIP=ON \
123
- -DCLR_BUILD_OCL=OFF \
124
- -DHIP_PLATFORM=amd \
125
- .. \
126
- && make -j$(nproc) \
127
- && make install \
128
- && rm -rf /tmp/rocm-systems
107
108
###
109
### Triton Build
@@ -259,9 +237,8 @@ ARG AITER_REPO
237
ARG USE_SCCACHE
238
RUN --mount=type=bind,from=build_pytorch,src=/app/install/,target=/install \
239
pip install /install/*.whl
262
-RUN git clone --recursive ${AITER_REPO}
240
+RUN git clone --recursive --branch ${AITER_BRANCH} ${AITER_REPO}
241
RUN cd aiter \
264
- && git checkout ${AITER_BRANCH} \
242
&& git submodule update --init --recursive \
243
&& pip install -r requirements.txt
244
RUN pip install pyyaml && cd aiter \