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 4, 07:30 UTC
[ROCm] ROCm7.2.2 + profiler fix + AITER 0.1.12.post2 (#41386)
Signed-off-by: Rohan138 <rohanpotdar138@gmail.com> Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com> Co-authored-by: Rohan138 <rohanpotdar138@gmail.com>
Gregory Shtrasberg committed
May 4, 2026 at 13:07 UTC
e724b0ea8d3b34e349a8635b1b49cb3005bf25f5
2 files changed
+25
-5
.buildkite/release-pipeline.yaml
+1
-1
@@ -723,7 +723,7 @@ steps:
723
- "bash tools/vllm-rocm/generate-rocm-wheels-root-index.sh"
724
env:
725
S3_BUCKET: "vllm-wheels"
726
- VARIANT: "rocm721"
726
+ VARIANT: "rocm722"
727
728
# ROCm Job 6: Build ROCm Release Docker Image
729
- label: ":docker: Build release image - x86_64 - ROCm"
docker/Dockerfile.rocm_base
+24
-4
@@ -1,4 +1,4 @@
1
-ARG BASE_IMAGE=rocm/dev-ubuntu-22.04:7.2.1-complete
1
+ARG BASE_IMAGE=rocm/dev-ubuntu-22.04:7.2.2-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
@@ -9,7 +9,7 @@ ARG PYTORCH_AUDIO_BRANCH="v2.9.0"
9
ARG PYTORCH_AUDIO_REPO="https://github.com/pytorch/audio.git"
10
ARG FA_BRANCH="0e60e394"
11
ARG FA_REPO="https://github.com/Dao-AILab/flash-attention.git"
12
-ARG AITER_BRANCH="v0.1.10.post3"
12
+ARG AITER_BRANCH="v0.1.12.post2"
13
ARG AITER_REPO="https://github.com/ROCm/aiter.git"
14
ARG MORI_BRANCH="v1.1.0"
15
ARG MORI_REPO="https://github.com/ROCm/mori.git"
@@ -104,6 +104,28 @@ 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
129
130
###
131
### Triton Build
@@ -153,8 +175,6 @@ RUN git clone ${PYTORCH_REPO} pytorch
175
RUN cd pytorch && git checkout ${PYTORCH_BRANCH}
176
RUN cd pytorch \
177
&& pip install -r requirements.txt && git submodule update --init --recursive
156
-RUN cd pytorch/third_party/kineto \
157
- && git remote add rocm https://github.com/ROCm/kineto && git fetch rocm && git checkout 2d73be3
178
RUN cd pytorch && python3 tools/amd_build/build_amd.py \
179
&& if [ "$USE_SCCACHE" = "1" ]; then \
180
export HIP_CLANG_PATH=/opt/sccache-wrappers \