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 17, 01:45 UTC
[CI/Build] Adding functionality to reset the node's GPUs before processing. (#4213)
Alexei-V-Ivanov-AMD committed
Apr 25, 2024 at 11:37 UTC
7ee82bef1e71febc28757807f5df8191bb36d88e
1 file changed
+15
-1
.buildkite/run-amd-test.sh
+15
-1
@@ -5,6 +5,19 @@ set -ex
5
# Print ROCm version
6
rocminfo
7
8
+
9
+echo "reset" > /opt/amdgpu/etc/gpu_state
10
+
11
+while true; do
12
+ sleep 3
13
+ if grep -q clean /opt/amdgpu/etc/gpu_state; then
14
+ echo "GPUs state is \"clean\""
15
+ break
16
+ fi
17
+done
18
+
19
+
20
+
21
# Try building the docker image
22
docker build -t rocm -f Dockerfile.rocm .
23
@@ -14,7 +27,8 @@ trap remove_docker_container EXIT
27
remove_docker_container
28
29
# Run the image
17
-docker run --device /dev/kfd --device /dev/dri --network host --name rocm rocm python3 -m vllm.entrypoints.api_server &
30
+export HIP_VISIBLE_DEVICES=1
31
+docker run --device /dev/kfd --device /dev/dri --network host -e HIP_VISIBLE_DEVICES --name rocm rocm python3 -m vllm.entrypoints.api_server &
32
33
# Wait for the server to start
34
wait_for_server_to_start() {