-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
179 lines (151 loc) · 8.35 KB
/
main.py
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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
import os
import json
import shutil
from tqdm import tqdm
from pydantic import ValidationError
from datasets import load_dataset
from judges.cpp_judge import CppJudge
from judges.python_judge import PythonJudge
from judges.java_judge import JavaJudge
from utils.logger import Logger, JSONLogger
from utils.models import Problem, Config
from utils.utils import sanitize_filename
from providers.openai import OpenAIProvider
from providers.huggingface import HuggingFaceProvider
from providers.anthropic import AnthropicProvider
from providers.mistral import MistralProvider
from providers.google import GoogleProvider
def load_problems_from_hf(dataset_name: str, split: str = 'train') -> list[str]:
dataset = load_dataset(dataset_name, split=split)
return [json.dumps(problem) for problem in dataset]
def load_config(config_path: str) -> Config:
with open(config_path, 'r') as file:
config_json = json.load(file)
return Config(**config_json)
def generate_summary(results: list[dict]) -> str:
passed_count = sum(1 for result in results if result['pass'])
total_count = len(results)
return f"Passed {passed_count}/{total_count} test cases"
def load_existing_log(log_filename: str) -> dict:
if os.path.exists(log_filename):
with open(log_filename, 'r') as file:
return json.load(file)
return {}
def initialize_provider(config: Config, logger: Logger):
if config.provider == "openai":
return OpenAIProvider(config.api_key, config.model, config.base_prompt, logger, config.language)
elif config.provider == "huggingface":
return HuggingFaceProvider(config.model, config.base_prompt, logger, config.language)
elif config.provider == "anthropic":
return AnthropicProvider(config.api_key, config.model, config.base_prompt, logger, config.language)
elif config.provider == "mistral":
return MistralProvider(config.api_key, config.model, config.base_prompt, logger, config.language)
elif config.provider == "google":
return GoogleProvider(config.api_key, config.model, config.base_prompt, logger, config.language)
else:
logger.log('error', "Invalid provider specified")
raise ValueError("Invalid provider specified")
def initialize_judge(language: str, logger: Logger):
if language == "cpp":
return CppJudge(logger)
elif language == "python":
return PythonJudge(logger)
elif language == "java":
return JavaJudge(logger)
else:
logger.log('error', "Unsupported language specified")
raise ValueError("Unsupported language specified")
def process_problem(judge, provider, problem_data: dict, shots: int, ignore_time_limits: bool, json_logger: JSONLogger, logger: Logger, problems_passed: int, total_filtered_problems: int, index: int) -> int:
problem_title = problem_data['title']
sanitized_title = sanitize_filename(problem_title)
for shot in range(1, shots + 1):
solution = provider.generate_solution(problem_data)
if solution:
if isinstance(judge, JavaJudge):
try:
class_name = judge.get_class_name(solution)
source_file = os.path.join("temp", f"{class_name}.java")
binary_file = os.path.join("temp", f"{class_name}.class")
except ValueError as e:
logger.log('error', str(e))
json_logger.log_compilation_error(problem_data["title"], problem_data.get("category", "Uncategorized"), solution, str(e), problems_passed, shot)
continue
else:
source_file = os.path.join("temp", f"{sanitized_title}_shot_{shot}.{judge.language_extension}")
binary_file = os.path.join("temp", f"{sanitized_title}_binary_shot_{shot}")
with open(source_file, 'w') as file:
file.write(solution)
if isinstance(judge, PythonJudge):
compile_success = True
else:
compile_success = judge.compile_code(source_file, binary_file)
if compile_success:
try:
problem = Problem(**problem_data)
results = []
for test_case in problem.test_cases:
input_data = test_case.input
if isinstance(judge, PythonJudge):
result = judge.run_code(source_file, input_data, problem.time_limit, problem.memory_limit, ignore_time_limits)
elif isinstance(judge, JavaJudge):
result = judge.run_code(class_name, input_data, problem.time_limit, problem.memory_limit, ignore_time_limits)
else:
result = judge.run_code(binary_file, input_data, problem.time_limit, problem.memory_limit, ignore_time_limits)
result['pass'] = judge.validate_output(result['output'], test_case.output)
result['log'] = result.get('error', '') or ('Passed' if result['pass'] else 'Failed')
results.append(result)
summary = generate_summary(results)
logger.log('info', f"Problem {index + 1}/{total_filtered_problems} Shot {shot}: {summary}")
if all(result['pass'] for result in results):
problems_passed += 1
json_logger.log_problem(problem.title, problem.category or "Uncategorized", results, solution, problems_passed, {"shot": shot, "status": "passed"})
break
else:
json_logger.log_problem(problem.title, problem.category or "Uncategorized", results, solution, problems_passed, {"shot": shot, "status": "failed"})
except ValidationError as e:
logger.log('error', f"Problem validation error: {e}")
else:
logger.log('error', "Compilation failed")
json_logger.log_compilation_error(problem_data["title"], problem_data.get("category", "Uncategorized"), solution, "Compilation failed", problems_passed, shot)
else:
logger.log('error', "Solution generation failed")
json_logger.log_compilation_error(problem_data["title"], problem_data.get("category", "Uncategorized"), "No solution generated", "Solution generation failed", problems_passed, shot)
return problems_passed
def main():
logger = Logger()
config = load_config('config.json')
os.makedirs("benchmark", exist_ok=True)
os.makedirs("temp", exist_ok=True)
log_filename = os.path.join("benchmark", f"{sanitize_filename(config.provider)}_{sanitize_filename(config.model)}_{sanitize_filename(config.language)}_log.json")
if not config.continue_from_log:
if os.path.exists(log_filename):
os.remove(log_filename)
json_logger = JSONLogger(log_filename)
json_logger.log_initial_config(config)
else:
json_logger = JSONLogger(log_filename)
problems = load_problems_from_hf("juvi21/cses-fi-competitive-coding-problems")
categories_filter = config.categories
shots = config.shots
ignore_time_limits = config.ignore_time_limits
judge = initialize_judge(config.language, logger)
provider = initialize_provider(config, logger)
if categories_filter:
filtered_problems = [problem for problem in problems if json.loads(problem).get("category") in categories_filter]
else:
filtered_problems = problems
total_filtered_problems = len(filtered_problems)
problems_passed = json_logger.data.get("total_passed_problems", 0)
processed_titles = set(problem["title"] for problem in json_logger.data.get("problems", []))
for index, problem_str in enumerate(tqdm(filtered_problems, desc="Processing problems")):
problem_data = json.loads(problem_str)
problem_title = problem_data['title']
if problem_title in processed_titles:
logger.log('info', f"Skipping already processed problem: {problem_title}")
continue
logger.log('info', f"Judging problem: {problem_title}")
problems_passed = process_problem(judge, provider, problem_data, shots, ignore_time_limits, json_logger, logger, problems_passed, total_filtered_problems, index)
if os.path.exists("temp"):
shutil.rmtree("temp")
if __name__ == "__main__":
main()