Should not expect latency in result if errors is not null (#12)

* first commit

* refactor

Author: Bslabe123

benchmark_serving.py

@@ -476,25 +476,22 @@ async def benchmark(
         if res is None:
             continue
         latency, ttft, itl, errors = res
-        prompt_len, output_len, request_latency = latency
-        overall_results["latencies"].append(latency)
-        if ttft:
-            overall_results["ttfts"].append(ttft)
-            overall_results["tpots"].append((request_latency - ttft) / (output_len - 1) if output_len > 1 else 0)
-        if itl:
-            overall_results["itls"].extend(itl)
         if errors:
-            for k, v in errors.items():
-                overall_results["errors"][k] += v
-        per_model_results[chosen_model]["latencies"].append(latency)
-        if ttft:
-            per_model_results[chosen_model]["ttfts"].append(ttft)
-            per_model_results[chosen_model]["tpots"].append((request_latency - ttft) / (output_len - 1) if output_len > 1 else 0)
-        if itl:
-            per_model_results[chosen_model]["itls"].extend(itl)
-        if errors:
-            for k, v in errors.items():
-                per_model_results[chosen_model]["errors"][k] += v
+          for k, v in errors.items():
+              overall_results["errors"][k] += v
+              per_model_results[chosen_model]["errors"][k] += v
+        else:
+          prompt_len, output_len, request_latency = latency
+          overall_results["latencies"].append(latency)
+          per_model_results[chosen_model]["latencies"].append(latency)
+          if ttft:
+              overall_results["ttfts"].append(ttft)
+              overall_results["tpots"].append((request_latency - ttft) / (output_len - 1) if output_len > 1 else 0)
+              per_model_results[chosen_model]["ttfts"].append(ttft)
+              per_model_results[chosen_model]["tpots"].append((request_latency - ttft) / (output_len - 1) if output_len > 1 else 0)
+          if itl:
+              overall_results["itls"].extend(itl)     
+              per_model_results[chosen_model]["itls"].extend(itl)     
 
     benchmark_duration = time.time() - benchmark_start_time