A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix] Fix SLA tuner initialization (#27355)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Oct 23, 2025 at 11:43 UTC
6738e4a09325396305b525c30ab2f68923ed2eee
2 files changed
+4
-1
vllm/benchmarks/sweep/serve.py
+2
@@ -66,6 +66,8 @@ def run_benchmark(
66
):
67
benchmark_cmd = [
68
*bench_overrides.apply_to_cmd(bench_cmd),
69
+ "--percentile-metrics",
70
+ "ttft,tpot,itl,e2el",
71
"--save-result",
72
"--result-dir",
73
str(output_path.parent),
vllm/benchmarks/sweep/serve_sla.py
+2
-1
@@ -401,7 +401,8 @@ class SweepServeSLAArgs(SweepServeArgs):
401
402
@classmethod
403
def from_cli_args(cls, args: argparse.Namespace):
404
- base_args = super().from_cli_args(args)
404
+ # NOTE: Don't use super() as `from_cli_args` calls `cls()`
405
+ base_args = SweepServeArgs.from_cli_args(args)
406
407
if args.sla_params:
408
sla_params = SLASweep.read_json(args.sla_params)