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 11, 03:45 UTC
[ROCm][CI/Build] Update ROCm dockerfiles (#30991)
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Gregory Shtrasberg committed
Dec 19, 2025 at 21:19 UTC
0be149524c0e21d8a86496918b24a919ed7eb3aa
2 files changed
+11
-5
docker/Dockerfile.rocm
+6
@@ -130,6 +130,7 @@ RUN --mount=type=bind,from=export_vllm,src=/,target=/install \
130
&& uv pip install --system *.whl
131
132
ARG COMMON_WORKDIR
133
+ARG BASE_IMAGE
134
135
# Copy over the benchmark scripts as well
136
COPY --from=export_vllm /benchmarks ${COMMON_WORKDIR}/vllm/benchmarks
@@ -144,4 +145,9 @@ ENV SAFETENSORS_FAST_GPU=1
145
# Performance environment variable.
146
ENV HIP_FORCE_DEV_KERNARG=1
147
148
+# Workaround for ROCm profiler limits
149
+RUN echo "ROCTRACER_MAX_EVENTS=10000000" > ${COMMON_WORKDIR}/libkineto.conf
150
+ENV KINETO_CONFIG="${COMMON_WORKDIR}/libkineto.conf"
151
+RUN echo "VLLM_BASE_IMAGE=${BASE_IMAGE}" >> ${COMMON_WORKDIR}/versions.txt
152
+
153
CMD ["/bin/bash"]
docker/Dockerfile.rocm_base
+5
-5
@@ -1,15 +1,15 @@
1
-ARG BASE_IMAGE=rocm/dev-ubuntu-22.04:7.1-complete
2
-ARG TRITON_BRANCH="57c693b6"
1
+ARG BASE_IMAGE=rocm/dev-ubuntu-22.04:7.0-complete
2
+ARG TRITON_BRANCH="a272dfa8"
3
ARG TRITON_REPO="https://github.com/ROCm/triton.git"
4
-ARG PYTORCH_BRANCH="1c57644d"
5
-ARG PYTORCH_VISION_BRANCH="v0.23.0"
4
+ARG PYTORCH_BRANCH="89075173"
5
ARG PYTORCH_REPO="https://github.com/ROCm/pytorch.git"
6
+ARG PYTORCH_VISION_BRANCH="v0.24.1"
7
ARG PYTORCH_VISION_REPO="https://github.com/pytorch/vision.git"
8
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="59bd8ff2"
12
+ARG AITER_BRANCH="6af8b687"
13
ARG AITER_REPO="https://github.com/ROCm/aiter.git"
14
15
FROM ${BASE_IMAGE} AS base