xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-22 23:54:26 +08:00
Code Issues Packages Projects Releases Wiki Activity

[AMD] [P/D] Compute num gpus for ROCm correctly in run_accuracy_test.sh (#18568)

Browse Source 
Signed-off-by: Randall Smith <Randall.Smith@amd.com>
This commit is contained in:
rasmith 2025-05-22 20:45:35 -05:00 committed by GitHub
parent c32e249a23
commit 46791e1b4b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 11 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh
View File

@ -13,6 +13,8 @@ NUM_DECODE_INSTANCES=${NUM_DECODE_INSTANCES:-2} # Default to 2
# Find the git repository root directory # Find the git repository root directory
GIT_ROOT=$(git rev-parse --show-toplevel) GIT_ROOT=$(git rev-parse --show-toplevel)
SMI_BIN=$(which nvidia-smi || which rocm-smi)
# Trap the SIGINT signal (triggered by Ctrl+C) # Trap the SIGINT signal (triggered by Ctrl+C)
trap 'kill $(jobs -pr)' SIGINT SIGTERM EXIT trap 'kill $(jobs -pr)' SIGINT SIGTERM EXIT
@ -44,6 +46,13 @@ get_model_args() {
echo "$extra_args" echo "$extra_args"
} }
get_num_gpus() {
if [[ "$SMI_BIN" == *"nvidia"* ]]; then
echo "$($SMI_BIN --query-gpu=name --format=csv,noheader | wc -l)"
else
echo "$($SMI_BIN -l | grep GPU | wc -l)"
fi
}
# Function to run tests for a specific model # Function to run tests for a specific model
run_tests_for_model() { run_tests_for_model() {
@ -64,7 +73,7 @@ run_tests_for_model() {
# Start prefill instances # Start prefill instances
for i in $(seq 0 $((NUM_PREFILL_INSTANCES-1))); do for i in $(seq 0 $((NUM_PREFILL_INSTANCES-1))); do
# Calculate GPU ID - we'll distribute across available GPUs # Calculate GPU ID - we'll distribute across available GPUs
GPU_ID=$((i % $(nvidia-smi --query-gpu=name --format=csv,noheader | wc -l))) GPU_ID=$((i % $(get_num_gpus)))
# Calculate port number (base port + instance number) # Calculate port number (base port + instance number)
PORT=$((8100 + i)) PORT=$((8100 + i))
# Calculate side channel port # Calculate side channel port
@ -96,7 +105,7 @@ run_tests_for_model() {
# Start decode instances # Start decode instances
for i in $(seq 0 $((NUM_DECODE_INSTANCES-1))); do for i in $(seq 0 $((NUM_DECODE_INSTANCES-1))); do
# Calculate GPU ID - we'll distribute across available GPUs, starting from after prefill GPUs # Calculate GPU ID - we'll distribute across available GPUs, starting from after prefill GPUs
GPU_ID=$(((i + NUM_PREFILL_INSTANCES) % $(nvidia-smi --query-gpu=name --format=csv,noheader | wc -l))) GPU_ID=$(((i + NUM_PREFILL_INSTANCES) % $(get_num_gpus)))
# Calculate port number (base port + instance number) # Calculate port number (base port + instance number)
PORT=$((8200 + i)) PORT=$((8200 + i))
# Calculate side channel port # Calculate side channel port