[ci] Remove --runtime=nvidia from launch_container.sh

Author: ethnzhng

tests/integration/launch_container.sh

@@ -84,10 +84,7 @@ if [[ "$(support_nvme)" == *"true"* ]]; then
 fi
 
 is_llm=false
-if [[ "$platform" == *"-gpu"* ]]; then # if the platform has cuda capabilities
-  runtime="nvidia"
 elif [[ "$platform" == *"lmi"* || "$platform" == *"trtllm"* || "$platform" == *"tensorrt-llm"* ]]; then # Runs multi-gpu
-  runtime="nvidia"
   is_llm=true
   if [[ "$(is_p4d_or_p5)" == *"true"* || $is_multi_node ]]; then
     shm="20gb"
@@ -157,7 +154,6 @@ if $is_multi_node; then
     -e DJL_LEADER_ADDR=${leader_hostname} \
     -e DJL_WORKER_ADDR_FORMAT="${LWS_NAME}-${GROUP_INDEX}-%d.${LWS_NAME}.${NAMESPACE}" \
     ${env_file} \
-    ${runtime:+--runtime="${runtime}"} \
     ${shm:+--shm-size="${shm}"} \
     ${host_device:+ ${host_device}} \
     "${docker_image}" "service ssh start; djl-serving"
@@ -179,7 +175,6 @@ if $is_multi_node; then
     -e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
     -e AWS_SESSION_TOKEN=$AWS_SESSION_TOKEN \
     ${env_file} \
-    ${runtime:+--runtime="${runtime}"} \
     ${shm:+--shm-size="${shm}"} \
     ${host_device:+ ${host_device}} \
     "${docker_image}" "service ssh start; /usr/bin/python3 /opt/djl/partition/run_multi_node_setup.py 2>&1 | tee /opt/djl/logs/lmi-worker.log; tail -f"
@@ -207,7 +202,6 @@ elif $is_sm_neo_context; then
     -e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
     -e AWS_SESSION_TOKEN=$AWS_SESSION_TOKEN \
     ${env_file} \
-    ${runtime:+--runtime="${runtime}"} \
     ${shm:+--shm-size="${shm}"} \
     ${host_device:+ ${host_device}} \
     "${docker_image}"
@@ -228,7 +222,6 @@ elif $is_partition; then
     -e AWS_ACCESS_KEY_ID=$AWS_ACCESS_KEY_ID \
     -e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
     -e AWS_SESSION_TOKEN=$AWS_SESSION_TOKEN \
-    ${runtime:+--runtime="${runtime}"} \
     ${shm:+--shm-size="${shm}"} \
     ${host_device:+ ${host_device}} \
     "${docker_image}" \
@@ -247,7 +240,6 @@ elif [[ "$docker_image" == *"text-generation-inference"* ]]; then
     -e AWS_ACCESS_KEY_ID=$AWS_ACCESS_KEY_ID \
     -e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
     -e AWS_SESSION_TOKEN=$AWS_SESSION_TOKEN \
-    ${runtime:+--runtime="${runtime}"} \
     ${shm:+--shm-size="${shm}"} \
     "${docker_image}" \
     ${args})
@@ -272,7 +264,6 @@ else
     -e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY \
     -e AWS_SESSION_TOKEN=$AWS_SESSION_TOKEN \
     $uid_mapping \
-    ${runtime:+--runtime="${runtime}"} \
     ${shm:+--shm-size="${shm}"} \
     ${host_device:+ ${host_device}} \
     "${docker_image}" \

tests/integration/llm/prepare.py

@@ -458,6 +458,7 @@
         "option.trust_remote_code": True,
         "option.max_model_len": 16384,
         "option.gpu_memory_utilization": "0.9",
+        "option.enable_expert_parallel": True,
     },
     "llama-4-scout-17b-16e-instruct": {
         "option.model_id": "s3://djl-llm/Llama-4-Scout-17B-16E-Instruct/",