chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

fix: use consistent time denominator for throughput metrics in bench_one_batch_server (#19223)

Browse Source
This commit is contained in:
Ajay Anubolu
2026-03-05 15:58:17 -08:00
committed by GitHub
parent dd2bbe6d62
commit 13af7cbb02
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/test/bench_one_batch_server_internal.py
View File

@@ -600,8 +600,8 @@ def run_one_case(
# Compute metrics
latency = time.perf_counter() - tic
input_throughput = batch_size * input_len / last_ttft
output_throughput = batch_size * output_len / (latency - last_ttft)
input_throughput = batch_size * input_len / latency
output_throughput = batch_size * output_len / latency
overall_throughput = batch_size * (input_len + output_len) / latency
if backend == "vllm":