1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144"""
Grader Test Runner for OpenJudge
This script evaluates the performance of a specified grader (e.g., ToolCallAccuracyGrader)
on a set of test cases from the BFCL dataset. It supports random sampling, skipping initial
cases, and validating scores against expected min/max thresholds.
Usage:
python run_grader_tests.py --grader_name <GraderClassName> --sample_size <N> --skip <K>
Environment Variables:
- BASE_URL: The base URL for the LLM API endpoint.
- API_KEY: The API key for authenticating with the LLM service.
Example:
python run_grader_eval_bfcl_dataset.py --grader_name ToolCallAccuracyGrader --sample_size 50 --skip 10
"""
import argparse
import asyncio
import json
import os
import random
from typing import List
import nest_asyncio
from openjudge.graders.agent import *
from openjudge.graders.common import *
from openjudge.models.schema.prompt_template import LanguageEnum
nest_asyncio.apply()
def read_jsonl(file_path: str) -> List[dict]:
with open(file_path, "r", encoding="utf-8") as f:
return [json.loads(line) for line in f if line.strip()]
def random_sample_jsonl(data, sample_size, seed=42):
if not sample_size:
print(f"All data will be selected because sample_size is not set")
return data
if sample_size > len(data):
raise ValueError(f"Random select {sample_size} samples, but only {len(data)} samples exists")
random.seed(seed)
sample_data = random.sample(data, sample_size)
return sample_data
def get_required_env(var_name):
value = os.getenv(var_name)
if not value:
raise EnvironmentError(f"Required environment variable '{var_name}' is not set.")
return value
def run_cases(case_file: str, grader_name: str, sample_size: int, skip: int):
model_config = {
"model": "qwen-long",
"base_url": get_required_env("BASE_URL"),
"api_key": get_required_env("API_KEY"),
"temperature": 0,
"stream": False,
}
cnt = 0
cases = read_jsonl(case_file)
select_cases = cases[skip:]
sample_data = random_sample_jsonl(select_cases, sample_size)
num_data = len(sample_data)
for data in sample_data:
cls_name = grader_name
kwargs = data.get("parameters")
index = data["index"]
try:
grader = eval(cls_name)(model=model_config, language=LanguageEnum.ZH)
result = asyncio.run(grader.aevaluate(**kwargs))
has_min = "min_expect_score" in data
has_max = "max_expect_score" in data
if not (has_min or has_max):
print(f"\033[91mFAILED \033[0m: index: {index}, missing min_expect_score or max_expect_score")
continue
if has_min and result.score < data["min_expect_score"]:
print(
f"\033[91mFAILED \033[0m, index: {index}, result score {result.score} is less than min_expect_score {data['min_expect_score']}, result: {result}"
)
continue
if has_max and result.score > data["max_expect_score"]:
print(
f"\033[91mFAILED\033[0m, index: {index}, result score {result.score} is greater than max_expect_score {data['max_expect_score']}, result: {result}"
)
continue
cnt += 1
print(f"PASSED: index: {index}, score: {result.score}")
except Exception as e:
print(f"failed case index: {index}")
print(e)
pass_rate = cnt / num_data
print(f"{grader_name} pass rate: {pass_rate} ({cnt}/{num_data})")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Run grader test cases from BFCL dataset.")
parser.add_argument(
"--case_file",
type=str,
default="bfcl_v3/tool_call/tool_call_bfcl_v3_multiple_eval_data.jsonl",
help="Specify the case file as input data",
)
parser.add_argument(
"--grader_name",
type=str,
default="ToolCallAccuracyGrader",
help="Specify the name of grader to run",
)
parser.add_argument(
"--sample_size",
type=int,
default=0,
help="Number of random sampling",
)
parser.add_argument(
"--skip",
type=int,
default=0,
help="Number of test cases to skip from the beginning (default: 0)",
)
args = parser.parse_args()
run_cases(
case_file=args.case_file,
grader_name=args.grader_name,
sample_size=args.sample_size,
skip=args.skip,
)