Update L0_multi_gpu

triton-inference-server · Oct 20, 2023 · d725497 · d725497
1 parent 4598bbc
commit d725497
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/ci/L0_multi_gpu/test.sh b/ci/L0_multi_gpu/test.sh
@@ -30,15 +30,15 @@ source ../common/util.sh
 TRITON_DIR=${TRITON_DIR:="/opt/tritonserver"}
 SERVER=${TRITON_DIR}/bin/tritonserver
 BACKEND_DIR=${TRITON_DIR}/backends
-SERVER_ARGS="--model-repository=`pwd`/models --backend-directory=${BACKEND_DIR} --log-verbose=1"
+SERVER_ARGS="--model-repository=`pwd`/models --backend-directory=${BACKEND_DIR} --model-control-mode=explicit --log-verbose=1"
 SERVER_LOG="./vllm_multi_gpu_test_server.log"
 CLIENT_LOG="./vllm_multi_gpu_test_client.log"
 TEST_RESULT_FILE='test_results.txt'
 CLIENT_PY="./vllm_multi_gpu_test.py"
 EXPECTED_NUM_TESTS=1
 
 mkdir -p models/vllm_opt/1/
-echo '{"model":"facebook/opt-125m", "disable_log_requests": "true", "gpu_memory_utilization":0.5, "tensor_parallel_size":2}' > models/vllm_opt/1/model.json
+echo '{"model":"facebook/opt-125m", "disable_log_requests": "true", "gpu_memory_utilization":0.25, "tensor_parallel_size":2}' > models/vllm_opt/1/model.json
 cp ../qa_models/vllm_opt/config.pbtxt models/vllm_opt
 
 pip3 install tritonclient

diff --git a/ci/L0_multi_gpu/vllm_multi_gpu_test.py b/ci/L0_multi_gpu/vllm_multi_gpu_test.py
@@ -59,7 +59,7 @@ def test_vllm_multi_gpu_utilization(self):
         gpu_ids = self.get_available_gpu_ids()
         self.assertGreaterEqual(len(gpu_ids), 2, "Error: Detected single GPU")
 
-        print("=============== Before Loading vLLM Model ===============")
+        print("\n\n=============== Before Loading vLLM Model ===============")
         mem_util_before_loading_model = {}
         for gpu_id in gpu_ids:
             memory_utilization = self.get_gpu_memory_utilization(gpu_id)