[Benchmark] Add an opt-out for the token-capacity check (#39284)

This commit is contained in:
Xingyu Liu
2026-09-14 17:17:12 -07:00
committed by GitHub
parent c0b8725f5a
commit 8874c51a96
@@ -124,6 +124,7 @@ class BenchArgs:
base_url: str = "" base_url: str = ""
local_tokenizer_path: str = "" local_tokenizer_path: str = ""
skip_warmup: bool = False skip_warmup: bool = False
skip_token_capacity_check: bool = False
show_report: bool = False show_report: bool = False
profile: bool = False profile: bool = False
profile_activities: Tuple[str] = ("CPU", "GPU") profile_activities: Tuple[str] = ("CPU", "GPU")
@@ -207,6 +208,11 @@ class BenchArgs:
), ),
) )
parser.add_argument("--skip-warmup", action="store_true") parser.add_argument("--skip-warmup", action="store_true")
parser.add_argument(
"--skip-token-capacity-check",
action="store_true",
help="Skip the raw-token capacity check; keep max-running-requests checks.",
)
parser.add_argument("--show-report", action="store_true") parser.add_argument("--show-report", action="store_true")
parser.add_argument("--profile", action="store_true") parser.add_argument("--profile", action="store_true")
parser.add_argument( parser.add_argument(
@@ -1230,6 +1236,9 @@ def run_benchmark_internal(
"token_capacity", 1000000000 "token_capacity", 1000000000
) )
if bench_args.skip_token_capacity_check:
skip_token_capacity_threshold = float("inf")
# Router /get_server_info responses carry "router_manager"; worker # Router /get_server_info responses carry "router_manager"; worker
# responses never do, so its presence confirms a router by design. # responses never do, so its presence confirms a router by design.
if not internal_states and server_info.get("router_manager"): if not internal_states and server_info.get("router_manager"):