ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

[fix] vllm-online-benchmark first token latency error (#12271)

Browse source
This commit is contained in:
Jun Wang 2024-10-29 17:54:36 +08:00 committed by GitHub
parent 0bbc04b5ec
commit 3700e81977
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 1 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
docker/llm/serving/xpu/docker/vllm_online_benchmark.py
View file

@ -270,13 +270,7 @@ def perform_request(session, url, payload, headers):
json_data = json.loads(data) json_data = json.loads(data)
if 'choices' in json_data and len(json_data['choices']) > 0: if 'choices' in json_data and len(json_data['choices']) > 0:
choice = json_data['choices'][0] choice = json_data['choices'][0]
if 'finish_reason' in choice and (choice['finish_reason'] == 'length' or choice['finish_reason'] == 'stop'): if 'text' in choice:
if 'first_token_time' in choice and isinstance(choice['first_token_time'], float):
first_token_inference_time = choice['first_token_time']
if 'rest_token_time' in choice and isinstance(choice['rest_token_time'], float):
next_token_inference_time = choice['rest_token_time']
else:
# 记录第一个token的时间
if first_token_time is None: if first_token_time is None:
first_token_time = token_time first_token_time = token_time
else: else: