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 15, 17:15 UTC
Applying some fixes for K8s agents in CI (#15493)
Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com>
Alexei-V-Ivanov-AMD committed
Mar 26, 2025 at 15:35 UTC
dd8a29da99aaca4aaedf710c813222871245e140
2 files changed
+8
-5
.buildkite/run-amd-test.sh
+6
-4
@@ -134,9 +134,10 @@ if [[ $commands == *"--shard-id="* ]]; then
134
# assign shard-id for each shard
135
commands_gpu=${commands//"--shard-id= "/"--shard-id=${GPU} "}
136
echo "Shard ${GPU} commands:$commands_gpu"
137
+ echo "Render devices: $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES"
138
docker run \
138
- --device /dev/kfd --device /dev/dri \
139
- --network host \
139
+ --device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
140
+ --network=host \
141
--shm-size=16gb \
142
--rm \
143
-e HIP_VISIBLE_DEVICES="${GPU}" \
@@ -163,9 +164,10 @@ if [[ $commands == *"--shard-id="* ]]; then
164
fi
165
done
166
else
167
+ echo "Render devices: $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES"
168
docker run \
167
- --device /dev/kfd --device /dev/dri \
168
- --network host \
169
+ --device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
170
+ --network=host \
171
--shm-size=16gb \
172
--rm \
173
-e HIP_VISIBLE_DEVICES=0 \
Dockerfile.rocm
+2
-1
@@ -12,7 +12,8 @@ ENV PYTORCH_ROCM_ARCH=${ARG_PYTORCH_ROCM_ARCH:-${PYTORCH_ROCM_ARCH}}
12
13
# Install some basic utilities
14
RUN apt-get update -q -y && apt-get install -q -y \
15
- sqlite3 libsqlite3-dev libfmt-dev libmsgpack-dev libsuitesparse-dev
15
+ sqlite3 libsqlite3-dev libfmt-dev libmsgpack-dev libsuitesparse-dev \
16
+ apt-transport-https ca-certificates wget curl
17
# Remove sccache
18
RUN python3 -m pip install --upgrade pip && pip install setuptools_scm
19
RUN apt-get purge -y sccache; python3 -m pip uninstall -y sccache; rm -f "$(which sccache)"