updating mi300 pt 2

Author: cquil11

runners/launch_mi300x-amd.sh

@@ -8,7 +8,7 @@ PORT=8888
 server_name="bmk-server"
 
 set -x
-docker run --rm --ipc=host --shm-size=16g --name=$server_name \
+docker run --rm --ipc=host --shm-size=16g --network=host --name=$server_name \
 --privileged --cap-add=CAP_SYS_ADMIN --device=/dev/kfd --device=/dev/dri --device=/dev/mem \
 --cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
 -v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \

runners/launch_mi300x-cr.sh

@@ -5,52 +5,16 @@ sudo sh -c 'echo 0 > /proc/sys/kernel/numa_balancing'
 HF_HUB_CACHE_MOUNT="/mnt/vdb/gha_cache/hf_hub_cache/"
 PORT=8888
 
-network_name="bmk-net"
 server_name="bmk-server"
-client_name="bmk-client"
-
-docker network create $network_name
 
 set -x
-docker run --rm -d --ipc=host --shm-size=16g --network=$network_name --name=$server_name \
+docker run --rm -d --ipc=host --shm-size=16g --network=host --name=$server_name \
 --privileged --cap-add=CAP_SYS_ADMIN --device=/dev/kfd --device=/dev/dri --device=/dev/mem \
 --cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
 -v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \
 -v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
 -e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e PORT=$PORT \
--e ISL -e OSL \
+-e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME \
 --entrypoint=/bin/bash \
 $IMAGE \
 benchmarks/"${EXP_NAME%%_*}_${PRECISION}_mi300x_docker.sh"
-
-set +x
-while IFS= read -r line; do
-    printf '%s\n' "$line"
-    if [[ "$line" =~ Application\ startup\ complete ]]; then
-        break
-    fi
-done < <(docker logs -f --tail=0 $server_name 2>&1)
-
-git clone https://github.com/kimbochen/bench_serving.git
-
-set -x
-docker run --rm --network=$network_name --name=$client_name \
--v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
--e HF_TOKEN -e PYTHONPYCACHEPREFIX=/tmp/pycache/ \
---entrypoint=python3 \
-$IMAGE \
-bench_serving/benchmark_serving.py \
---model=$MODEL --backend=vllm --base-url=http://$server_name:$PORT \
---dataset-name=random \
---random-input-len=$ISL --random-output-len=$OSL --random-range-ratio=$RANDOM_RANGE_RATIO \
---num-prompts=$(( $CONC * 10 )) \
---max-concurrency=$CONC \
---request-rate=inf --ignore-eos \
---save-result --percentile-metrics="ttft,tpot,itl,e2el" \
---result-dir=/workspace/ --result-filename=$RESULT_FILENAME.json
-
-while [ -n "$(docker ps -aq)" ]; do
-    docker stop $server_name
-    docker network rm $network_name
-    sleep 5
-done

runners/launch_mi300x-oci.sh

@@ -3,52 +3,16 @@
 HF_HUB_CACHE_MOUNT="$HOME/hf_hub_cache/"
 PORT=8888
 
-network_name="bmk-net"
 server_name="bmk-server"
-client_name="bmk-client"
-
-docker network create $network_name
 
 set -x
-docker run --rm -d --ipc=host --shm-size=16g --network=$network_name --name=$server_name \
+docker run --rm -d --ipc=host --shm-size=16g --network=host --name=$server_name \
 --privileged --cap-add=CAP_SYS_ADMIN --device=/dev/kfd --device=/dev/dri --device=/dev/mem \
 --cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
 -v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \
 -v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
 -e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e PORT=$PORT \
--e ISL -e OSL \
+-e ISL -e OSL -e PYTHONPYCACHEPREFIX=/tmp/pycache/ -e RANDOM_RANGE_RATIO -e RESULT_FILENAME \
 --entrypoint=/bin/bash \
 $IMAGE \
 benchmarks/"${EXP_NAME%%_*}_${PRECISION}_mi300x_docker.sh"
-
-set +x
-while IFS= read -r line; do
-    printf '%s\n' "$line"
-    if [[ "$line" =~ Application\ startup\ complete ]]; then
-        break
-    fi
-done < <(docker logs -f --tail=0 $server_name 2>&1)
-
-git clone https://github.com/kimbochen/bench_serving.git
-
-set -x
-docker run --rm --network=$network_name --name=$client_name \
--v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
--e HF_TOKEN -e PYTHONPYCACHEPREFIX=/tmp/pycache/ \
---entrypoint=python3 \
-$IMAGE \
-bench_serving/benchmark_serving.py \
---model=$MODEL --backend=vllm --base-url=http://$server_name:$PORT \
---dataset-name=random \
---random-input-len=$ISL --random-output-len=$OSL --random-range-ratio=$RANDOM_RANGE_RATIO \
---num-prompts=$(( $CONC * 10 )) \
---max-concurrency=$CONC \
---request-rate=inf --ignore-eos \
---save-result --percentile-metrics="ttft,tpot,itl,e2el" \
---result-dir=/workspace/ --result-filename=$RESULT_FILENAME.json
-
-while [ -n "$(docker ps -aq)" ]; do
-    docker stop $server_name
-    docker network rm $network_name
-    sleep 5
-done