From b23477af4433029d1e5621d68eb35c07e2a038bc Mon Sep 17 00:00:00 2001 From: shuwenn <47200617+alphabetc1@users.noreply.github.com> Date: Tue, 16 Jun 2026 11:03:12 +0800 Subject: [PATCH] bench: infer tokenizer from serving model info (#28195) --- python/sglang/bench_serving.py | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-) diff --git a/python/sglang/bench_serving.py b/python/sglang/bench_serving.py index 309997a71..a11adfa22 100644 --- a/python/sglang/bench_serving.py +++ b/python/sglang/bench_serving.py @@ -1931,7 +1931,20 @@ def run_benchmark(args_: argparse.Namespace): # Read dataset backend = args.backend model_id = args.served_model_name or args.model - tokenizer_id = args.tokenizer if args.tokenizer is not None else args.model + tokenizer_id = args.tokenizer + if tokenizer_id is None: + try: + resp = requests.get( + base_url + "/model_info", headers=get_auth_headers(), timeout=5 + ) + if resp.status_code == 200: + info = resp.json() + tokenizer_id = info.get("tokenizer_path") or info.get("model_path") + except Exception: + pass + if tokenizer_id is None: + tokenizer_id = args.model + tokenizer = get_tokenizer(tokenizer_id) input_requests = get_dataset(args, tokenizer, model_id)