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 5, 13:30 UTC
Removed GPU state confirmation and cleanup steps. (#38238)
Signed-off-by: Divin Honnappa <divin.honnappa@amd.com>
Divin Honnappa committed
Apr 3, 2026 at 15:11 UTC
121ea5a21f0d16e63e98883e17489927adea0728
1 file changed
-24
.buildkite/scripts/hardware_ci/run-amd-test.sh
-24
@@ -35,23 +35,6 @@ export PYTHONPATH=".."
35
# Helper Functions
36
###############################################################################
37
38
-wait_for_clean_gpus() {
39
- local timeout=${1:-300}
40
- local start=$SECONDS
41
- echo "--- Waiting for clean GPU state (timeout: ${timeout}s)"
42
- while true; do
43
- if grep -q clean /opt/amdgpu/etc/gpu_state; then
44
- echo "GPUs state is \"clean\""
45
- return
46
- fi
47
- if (( SECONDS - start >= timeout )); then
48
- echo "Error: GPUs did not reach clean state within ${timeout}s" >&2
49
- exit 1
50
- fi
51
- sleep 3
52
- done
53
-}
54
-
38
cleanup_docker() {
39
# Get Docker's root directory
40
docker_root=$(docker info -f '{{.DockerRootDir}}')
@@ -365,19 +348,12 @@ apply_rocm_test_overrides() {
348
###############################################################################
349
350
# --- GPU initialization ---
368
-echo "--- Confirming Clean Initial State"
369
-wait_for_clean_gpus
370
-
351
echo "--- ROCm info"
352
rocminfo
353
354
# --- Docker housekeeping ---
355
cleanup_docker
356
377
-echo "--- Resetting GPUs"
378
-echo "reset" > /opt/amdgpu/etc/gpu_state
379
-wait_for_clean_gpus
380
-
357
# --- Pull test image ---
358
echo "--- Pulling container"
359
image_name="rocm/vllm-ci:${BUILDKITE_COMMIT}"