From 9afc51e5b2cf120f4f7d72e82a884068cc4c696c Mon Sep 17 00:00:00 2001 From: Hyunjae Woo Date: Tue, 7 Nov 2023 16:31:59 -0800 Subject: [PATCH] Do not reset token-to-token latencies list --- src/c++/perf_analyzer/docs/examples/profile.py | 1 - 1 file changed, 1 deletion(-) diff --git a/src/c++/perf_analyzer/docs/examples/profile.py b/src/c++/perf_analyzer/docs/examples/profile.py index 958961ee4..534dcec95 100644 --- a/src/c++/perf_analyzer/docs/examples/profile.py +++ b/src/c++/perf_analyzer/docs/examples/profile.py @@ -270,7 +270,6 @@ def collect_online_metrics(export_data, output_tokens): first_token_latencies.append(first_token_latency) generation_latencies.append(generation_latency_ms) generation_throughputs.append(output_tokens / generation_latency_s) - token_to_token_latencies = [] for prev_res, res in pairwise(responses): token_to_token_latencies.append((res - prev_res) / 1_000_000) return (