Fix duplicate flag addition and make sleep time configurable (#20)

* first commit

* dont remove licence

* revert todo

Author: Bslabe123

latency_throughput_curve.sh

@@ -13,6 +13,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 set -o xtrace
 
 export IP=$IP
@@ -24,48 +25,51 @@ if [[ "$PROMPT_DATASET" = "sharegpt" ]]; then
 fi
 
 PYTHON="python3"
-PYTHON_OPTS="benchmark_serving.py "
+BASE_PYTHON_OPTS=(
+  "benchmark_serving.py"
+  "--save-json-results"
+  "--host=$IP"
+  "--port=$PORT"
+  "--dataset=$PROMPT_DATASET_FILE"
+  "--tokenizer=$TOKENIZER"
+  "--backend=$BACKEND"
+  "--max-input-length=$INPUT_LENGTH"
+  "--max-output-length=$OUTPUT_LENGTH"
+  "--file-prefix=$FILE_PREFIX"
+  "--models=$MODELS"
+  "--pm-namespace=$PM_NAMESPACE"
+  "--pm-job=$PM_JOB"
+)
+
+[[ "$TRAFFIC_SPLIT" ]] && BASE_PYTHON_OPTS+=("--traffic-split=$TRAFFIC_SPLIT")
+[[ "$OUTPUT_BUCKET" ]] && BASE_PYTHON_OPTS+=("--output-bucket=$OUTPUT_BUCKET")
+[[ "$SCRAPE_SERVER_METRICS" = "true" ]] && BASE_PYTHON_OPTS+=("--scrape-server-metrics")
+[[ "$SAVE_AGGREGATED_RESULT" = "true" ]] && BASE_PYTHON_OPTS+=("--save-aggregated-result")
+[[ "$STREAM_REQUEST" = "true" ]] && BASE_PYTHON_OPTS+=("--stream-request")
+[[ "$OUTPUT_BUCKET_FILEPATH" ]] && BASE_PYTHON_OPTS+=("--output-bucket-filepath" "$OUTPUT_BUCKET_FILEPATH")
+
+SLEEP_TIME=${SLEEP_TIME:-0}
+
 for request_rate in $(echo $REQUEST_RATES | tr ',' ' '); do
   echo "Benchmarking request rate: ${request_rate}"
   # TODO: Check if profile already exists, if so then skip
   timestamp=$(date +"%Y-%m-%d_%H-%M-%S")
   output_file="latency-profile-${timestamp}.txt"
-  if [ ${request_rate} == 0 ]; then
+  
+  if [ "$request_rate" == "0" ]; then
     request_rate="inf"
     num_prompts=$MAX_NUM_PROMPTS
   else
     num_prompts=$(awk "BEGIN {print int($request_rate * $BENCHMARK_TIME_SECONDS)}")
   fi
-  
-  echo "TOTAL prompts: $num_prompts"
-  
-  # Build the python command options
-  PYTHON_OPTS="$PYTHON_OPTS --save-json-results --host=$IP --port=$PORT --dataset=$PROMPT_DATASET_FILE --tokenizer=$TOKENIZER --request-rate=$request_rate --backend=$BACKEND --num-prompts=$num_prompts --max-input-length=$INPUT_LENGTH --max-output-length=$OUTPUT_LENGTH --file-prefix=$FILE_PREFIX --models=$MODELS --pm-namespace=$PM_NAMESPACE --pm-job=$PM_JOB"
 
-  if [[ "$TRAFFIC_SPLIT" ]]; then
-    PYTHON_OPTS="$PYTHON_OPTS --traffic-split=$TRAFFIC_SPLIT"
-  fi
-
-  if [[ "$OUTPUT_BUCKET" ]]; then
-    PYTHON_OPTS="$PYTHON_OPTS --output-bucket=$OUTPUT_BUCKET"
-  fi
-
-  if [[ "$SCRAPE_SERVER_METRICS" = "true" ]]; then
-    PYTHON_OPTS="$PYTHON_OPTS --scrape-server-metrics"
-  fi
-  if [[ "$SAVE_AGGREGATED_RESULT" = "true" ]]; then
-    PYTHON_OPTS="$PYTHON_OPTS --save-aggregated-result"
-  fi
-  if [[ "$STREAM_REQUEST" = "true" ]]; then
-    PYTHON_OPTS="$PYTHON_OPTS --stream-request"
-  fi
-  if [[ "$OUTPUT_BUCKET_FILEPATH" ]]; then
-    PYTHON_OPTS="$PYTHON_OPTS --output-bucket-filepath $OUTPUT_BUCKET_FILEPATH"
-  fi
+  echo "TOTAL prompts: $num_prompts"
+  PYTHON_OPTS=("${BASE_PYTHON_OPTS[@]}" "--request-rate=$request_rate" "--num-prompts=$num_prompts")
 
-  $PYTHON $PYTHON_OPTS > $output_file
-  cat $output_file
-  sleep 30 # wait 30 seconds before next run to ensure metrics isolation
+  $PYTHON "${PYTHON_OPTS[@]}" > "$output_file"
+  cat "$output_file"
+  echo "Sleeping for $SLEEP_TIME seconds..."
+  sleep $SLEEP_TIME
 done
 
 export LPG_FINISHED="true"