Update L0_multi_gpu_vllm script in order to be able access resources within container. Those resources no longer shipped with 'pytorch-triton' wheel package.

Author: mc-nv

ci/L0_multi_gpu_vllm/multi_lora/test.sh

@@ -41,6 +41,9 @@ EXPECTED_NUM_TESTS=2
 GENERATE_ENDPOINT="localhost:8000/v2/models/vllm_llama_multi_lora/generate"
 CHECK_FOR_ERROR=true
 
+export C_INCLUDE_PATH=/usr/local/cuda/include:$C_INCLUDE_PATH
+export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
+
 make_api_call() {
     local endpoint="$1"
     local data="$2"

ci/L0_multi_gpu_vllm/vllm_backend/test.sh

@@ -36,6 +36,9 @@ CLIENT_PY="./vllm_multi_gpu_test.py"
 SAMPLE_MODELS_REPO="../../../samples/model_repository"
 EXPECTED_NUM_TESTS=1
 
+export C_INCLUDE_PATH=/usr/local/cuda/include:$C_INCLUDE_PATH
+export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
+
 ### Helpers
 function validate_file_contains() {
     local KEY="${1}"