Skip to content

Commit ce3b04e

Browse files
committed
update scripts/playground/replay_request_dump.py
1 parent e81e042 commit ce3b04e

File tree

1 file changed

+13
-6
lines changed

1 file changed

+13
-6
lines changed

scripts/playground/replay_request_dump.py

Lines changed: 13 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,10 @@
11
"""
22
Usage:
33
# replay from a folder
4-
python3 replay_request_dump.py --file-number 100 --parallel 512 --input-folder /data/folder
4+
python3 replay_request_dump.py --file-number 100 --parallel 512 --input-folder /data/lianmin/sglang_request_dump/grok-mini-0220-engine-5756f8f94-28bm6/
55
66
# replay from a single file
7-
python3 replay_request_dump.py --parallel 512 --input-file /data/file.pkl
7+
python3 replay_request_dump.py --parallel 512 --input-file /data/sglang_crash_dump/memx-cti-34-sr1.xpop.twttr.net/crash_dump_2025-06-04_20-13-18.pkl
88
"""
99

1010
import argparse
@@ -38,9 +38,19 @@ def run_one_request_internal(record):
3838
(req, output, replay_init_time, start_time, end_time, idx) = record
3939
time.sleep(max(0, start_time - (time.time() - replay_init_time)))
4040

41+
if "completion_tokens" in output.get("meta_info", {}):
42+
recorded_completion_tokens = output["meta_info"]["completion_tokens"]
43+
else:
44+
recorded_completion_tokens = ""
45+
4146
json_data = asdict(req)
4247
stream = json_data["stream"]
4348

49+
if args.ignore_eos:
50+
json_data["sampling_params"]["ignore_eos"] = True
51+
if recorded_completion_tokens:
52+
json_data["sampling_params"]["max_new_tokens"] = recorded_completion_tokens
53+
4454
response = requests.post(
4555
f"http://{args.host}:{args.port}/generate",
4656
json=json_data,
@@ -59,10 +69,6 @@ def run_one_request_internal(record):
5969

6070
prompt_tokens = ret["meta_info"]["prompt_tokens"]
6171
completion_tokens = ret["meta_info"]["completion_tokens"]
62-
if "completion_tokens" in ret["meta_info"]:
63-
recorded_completion_tokens = ret["meta_info"]["completion_tokens"]
64-
else:
65-
recorded_completion_tokens = ""
6672
print(
6773
f"{idx=}, {start_time=:.2f}, {prompt_tokens=}, "
6874
f"{completion_tokens=}, {recorded_completion_tokens=}"
@@ -114,6 +120,7 @@ def main(records):
114120
parser.add_argument("--req-start", type=int, default=0)
115121
parser.add_argument("--parallel", type=int, default=512)
116122
parser.add_argument("--idx", type=int, default=None)
123+
parser.add_argument("--ignore-eos", action="store_true")
117124
args = parser.parse_args()
118125

119126
set_ulimit()

0 commit comments

Comments
 (0)