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

bench: infer tokenizer from serving model info (#28195)

Browse Source
This commit is contained in:
shuwenn
2026-06-15 20:03:12 -07:00
committed by GitHub
parent 063ab89ac1
commit b23477af44
1 changed files with 14 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/bench_serving.py
+14 -1
View File
@@ -1931,7 +1931,20 @@ def run_benchmark(args_: argparse.Namespace):
# Read dataset # Read dataset
backend = args.backend backend = args.backend
model_id = args.served_model_name or args.model 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) tokenizer = get_tokenizer(tokenizer_id)
input_requests = get_dataset(args, tokenizer, model_id) input_requests = get_dataset(args, tokenizer, model_id)