[Benchmark] Download model from modelscope (#634)

### What this PR does / why we need it?
-  Run benchmark scripts will Download model from modelscope

Signed-off-by: wangli <wangli858794774@gmail.com>

benchmarks/scripts/run-performance-benchmarks.sh

@@ -264,6 +264,10 @@ main() {
   # turn of the reporting of the status of each request, to clean up the terminal output
   export VLLM_LOG_LEVEL="WARNING"
 
+  # set env
+  export VLLM_USE_MODELSCOPE="True"
+  export HF_ENDPOINT="https://hf-mirror.com"
+
   # prepare for benchmarking
   cd benchmarks || exit 1
   get_benchmarks_scripts

benchmarks/tests/latency-tests.json

@@ -2,7 +2,7 @@
   {
     "test_name": "latency_llama8B_tp1",
     "parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
+      "model": "LLM-Research/Meta-Llama-3.1-8B-Instruct",
       "tensor_parallel_size": 1,
       "load_format": "dummy",
       "num_iters_warmup": 5,

benchmarks/tests/serving-tests.json

@@ -8,7 +8,7 @@
       "inf"
     ],
     "server_parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
+      "model": "LLM-Research/Meta-Llama-3.1-8B-Instruct",
       "tensor_parallel_size": 1,
       "swap_space": 16,
       "disable_log_stats": "",
@@ -16,7 +16,7 @@
       "load_format": "dummy"
     },
     "client_parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
+      "model": "LLM-Research/Meta-Llama-3.1-8B-Instruct",
       "backend": "vllm",
       "dataset_name": "sharegpt",
       "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",

benchmarks/tests/throughput-tests.json

@@ -2,7 +2,7 @@
   {
     "test_name": "throughput_llama8B_tp1",
     "parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
+      "model": "LLM-Research/Meta-Llama-3.1-8B-Instruct",
       "tensor_parallel_size": 1,
       "load_format": "dummy",
       "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",