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

[misc] Unify bench seed default to 42 and rename --profile-filename-prefix to --profile-prefix (#28576)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-06-17 16:41:23 -07:00
committed by GitHub
parent 5ea0d1d093
commit b88bada64e
3 changed files with 16 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/bench_one_batch.py
+13 -11
View File
@@ -206,7 +206,7 @@ class BenchArgs:
profile_record_shapes: bool = False
profile_activities: Tuple[str] = ("CPU", "GPU")
profile_stage: str = "all"
profile_filename_prefix: str = "profile"
profile_prefix: str = "profile"
profile_start_step: Optional[int] = None
profile_steps: Optional[int] = None
@@ -258,11 +258,13 @@ class BenchArgs:
help="Which stage to profile: all, prefill, or decode only.",
)
parser.add_argument(
"--profile-filename-prefix",
"--profile-prefix",
"--profile-filename-prefix", # deprecated alias, kept for back-compat
dest="profile_prefix",
type=str,
default=BenchArgs.profile_filename_prefix,
default=BenchArgs.profile_prefix,
help="Prefix of the profiling file names. The full profiling result file(s) be "
'"[profile_filename_prefix]_batch[batch_size]_input[input_len]_output[output_len].trace.json.gz"',
'"[profile_prefix]_batch[batch_size]_input[input_len]_output[output_len].trace.json.gz"',
)
parser.add_argument(
"--profile-start-step",
@@ -612,10 +614,10 @@ def _get_torch_profiler_output_dir():
def _create_torch_profiler_filename(
profile_filename_prefix, batch_size, input_len, output_len, stage
profile_prefix, batch_size, input_len, output_len, stage
):
output_dir = _get_torch_profiler_output_dir()
filename = f"{profile_filename_prefix}_batch{batch_size}_input{input_len}_output{output_len}_{stage}.trace.json.gz"
filename = f"{profile_prefix}_batch{batch_size}_input{input_len}_output{output_len}_{stage}.trace.json.gz"
return os.path.join(output_dir, filename)
@@ -701,7 +703,7 @@ def latency_test_run_once(
profile,
profile_record_shapes,
profile_activities,
profile_filename_prefix,
profile_prefix,
profile_stage,
tp_rank,
profile_start_step=None,
@@ -730,7 +732,7 @@ def latency_test_run_once(
trace_filename_prefill = None
if enable_profile_prefill:
trace_filename_prefill = _create_torch_profiler_filename(
profile_filename_prefix, batch_size, input_len, output_len, "prefill"
profile_prefix, batch_size, input_len, output_len, "prefill"
)
profiler = start_profile(
profile_activities,
@@ -777,7 +779,7 @@ def latency_test_run_once(
# Start profiler at the specified step
if enable_profile_decode and i == profile_start:
trace_filename_decode = _create_torch_profiler_filename(
profile_filename_prefix, batch_size, input_len, output_len, "decode"
profile_prefix, batch_size, input_len, output_len, "decode"
)
profiler = start_profile(
profile_activities,
@@ -875,7 +877,7 @@ def latency_test(
profile=False,
profile_record_shapes=False,
profile_activities=("CPU", "GPU"),
profile_filename_prefix="",
profile_prefix="",
profile_stage="all",
tp_rank=tp_rank,
profile_start_step=None,
@@ -926,7 +928,7 @@ def latency_test(
bench_args.profile if tp_rank == 0 else None,
bench_args.profile_record_shapes if tp_rank == 0 else None,
bench_args.profile_activities,
bench_args.profile_filename_prefix,
bench_args.profile_prefix,
bench_args.profile_stage,
tp_rank,
bench_args.profile_start_step,