chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

ci: refactor nightly test (#10495)

Browse Source
This commit is contained in:
Mick
2025-09-27 06:24:30 +08:00
committed by GitHub
parent 05a3526654
commit 777eb53897
16 changed files with 1656 additions and 187 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
python/sglang/bench_one_batch.py
View File

@@ -443,11 +443,9 @@ def latency_test_run_once(
if profile:
profiler.stop()
profile_filename = f"{profile_filename_prefix}_batch{batch_size}_input{input_len}_output{output_len}_prefill.trace.json.gz"
_save_profile_trace_results(profiler, profile_filename)
rank_print(
f"torch profiler chrome trace for prefill saved to {profile_filename}"
)
trace_filename = f"{profile_filename_prefix}_batch{batch_size}_input{input_len}_output{output_len}_prefill.trace.json.gz"
_save_profile_trace_results(profiler, trace_filename)
rank_print(f"torch profiler chrome trace for prefill saved to {trace_filename}")
# Decode
decode_latencies = []
@@ -479,10 +477,10 @@ def latency_test_run_once(
if profile and i == output_len / 2:
profiler.stop()
profile_filename = f"{profile_filename_prefix}_batch{batch_size}_input{input_len}_output{output_len}_decode.trace.json.gz"
_save_profile_trace_results(profiler, profile_filename)
trace_filename = f"{profile_filename_prefix}_batch{batch_size}_input{input_len}_output{output_len}_decode.trace.json.gz"
_save_profile_trace_results(profiler, trace_filename)
rank_print(
f"torch profiler chrome trace for decoding 1 token saved to {profile_filename}"
f"torch profiler chrome trace for decoding 1 token saved to {trace_filename}"
)
# Record decode timing from 2nd output