Update benchmark_throughput.py

3_env_check&batches_llm_inference/scripts/benchmark_throughput.py

@@ -294,9 +294,9 @@ def run_vllm(
             info["ttft_median"] = np.around(np.median(total_ttfts or 0),5)
             info["ttft_p99"] = np.around(np.percentile(total_ttfts or 0, 99),5)
 
-            info["tpop_mean"] = np.around(np.mean(total_tpops),4)
-            info["tpop_median"] = np.around(np.median(total_tpops or 0),5)
-            info["tpop_p99"] = np.around(np.percentile(total_tpops or 0, 99),5)
+            info["tpot_mean"] = np.around(np.mean(total_tpops),4)
+            info["tpot_median"] = np.around(np.median(total_tpops or 0),5)
+            info["tpot_p99"] = np.around(np.percentile(total_tpops or 0, 99),5)
 
             info["output_token_throughput_mean"]  = np.around(np.mean(total_output_token_throughput),2)
             info["output_token_throughput_median"]  = np.around(np.median(total_output_token_throughput or 0),2)
@@ -319,9 +319,9 @@ def run_vllm(
             print(f"TTFT_mean: {info['ttft_mean']*1000: .2f} ms")       # 首字延时
             print(f"ttft_p99: {info['ttft_p99']*1000: .2f} ms")
             print(f"ttft_median: {info['ttft_median']*1000: .2f} ms")
-            print(f"TPOP_mean: {info['tpop_mean']*1000: .2f} ms")              # 单字decode时间
-            print(f"tpop_median: {info['tpop_median']*1000: .2f} ms")
-            print(f"tpop_p99: {info['tpop_p99']*1000: .2f} ms")
+            print(f"TPOT_mean: {info['tpop_mean']*1000: .2f} ms")              # 单字decode时间
+            print(f"tpot_median: {info['tpop_median']*1000: .2f} ms")
+            print(f"tpot_p99: {info['tpop_p99']*1000: .2f} ms")
             print(f"output_token_throughput_mean: {info['output_token_throughput_mean']:.2f} tokens/s")           # 单路生成吞吐
             print(f"output_token_throughput_median: {info['output_token_throughput_median']:.2f} tokens/s")
             print(f"output_token_throughput_p99: {info['output_token_throughput_p99']:.2f} tokens/s")