Fix python bool flag logic

GoogleCloudPlatform · Sep 12, 2024 · 19ffec5 · 19ffec5
1 parent d730ea3
commit 19ffec5
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 3 deletions.
diff --git a/benchmarks/benchmark/tools/profile-generator/container/benchmark_serving.py b/benchmarks/benchmark/tools/profile-generator/container/benchmark_serving.py
@@ -609,8 +609,7 @@ def main(args: argparse.Namespace):
   )
   parser.add_argument(
       "--scrape-server-metrics",
-      type=bool,
-      default=False,
+      action="store_true",
       help="Whether to scrape server metrics.",
   )
   cmd_args = parser.parse_args()

diff --git a/benchmarks/benchmark/tools/profile-generator/container/latency_throughput_curve.sh b/benchmarks/benchmark/tools/profile-generator/container/latency_throughput_curve.sh
@@ -19,11 +19,17 @@ export IP=$IP
 
 huggingface-cli login --token "$HF_TOKEN" --add-to-git-credential
 
+PYTHON="python3"
+PYTHON_OPTS="benchmark_serving.py "
 for request_rate in $(echo $REQUEST_RATES | tr ',' ' '); do
   # TODO: Check if profile already exists, if so then skip
   timestamp=$(date +"%Y-%m-%d_%H-%M-%S")
   output_file="latency-profile-${timestamp}.txt"
-  python3 benchmark_serving.py   --host="$IP"   --port="$PORT"   --model="$TOKENIZER" --dataset=ShareGPT_V3_unfiltered_cleaned_split.json   --tokenizer="$TOKENIZER" --request-rate=$request_rate --backend="$BACKEND" --num-prompts=$((request_rate * 30)) --max-input-length=$INPUT_LENGTH --max-output-length=$OUTPUT_LENGTH --scrape-server-metrics=$SCRAPE_SERVER_METRICS > $output_file
+  PYTHON_OPTS="$PYTHON_OPTS --host=$IP   --port=$PORT   --model=$TOKENIZER --dataset=ShareGPT_V3_unfiltered_cleaned_split.json --tokenizer=$TOKENIZER --request-rate=$request_rate --backend=$BACKEND --num-prompts=$((request_rate * 30)) --max-input-length=$INPUT_LENGTH --max-output-length=$OUTPUT_LENGTH"
+  if [[ "$SCRAPE_SERVER_METRICS" = "true" ]]; then
+    PYTHON_OPTS="$PYTHON_OPTS --scrape-server-metrics"
+  fi
+  $PYTHON $PYTHON_OPTS > $output_file
   cat $output_file
   sleep 5 # wait 5 seconds before next run
 done