File size: 8,451 Bytes
5fed0fc |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 |
import requests
import time
import os
import re
import json
import concurrent.futures
import argparse
import sys
import threading
from llm_interface import *
# --- Configuration ---
API_WORKERS = 16 # Max concurrent LLM generations
JUDGE_WORKERS = 8 # Max concurrent submissions to the local judge
# Get the directory where this script is currently located
script_dir = os.path.dirname(os.path.abspath(__file__))
# Set solution_dir to be a folder named 'solutions' inside that directory
solution_dir = os.path.join(script_dir, 'solutions')
# Define available models here for validation
AVAILABLE_MODELS = {
'gemini', 'gpt', 'claude', 'claude-opus',
'claude-sonnet-4-5', 'gemini3', 'Grok', 'claude-opus-4-5'
}
# Semaphore to control concurrent access to the judge
judge_throttler = threading.BoundedSemaphore(JUDGE_WORKERS)
class LocalJudge:
def __init__(self, judge_url="http://localhost:8081"):
self.judge_url = judge_url
self.session = requests.Session()
def get_all_problems(self):
try:
response = self.session.get(f"{self.judge_url}/problems")
response.raise_for_status()
return [p['id'] for p in response.json().get('problems', [])]
except requests.RequestException:
print(f"Error connecting to the judge at {self.judge_url}. Is it running?")
return None
def get_problem_statement(self, pid):
try:
response = self.session.get(f"{self.judge_url}/problem/{pid}/statement")
response.raise_for_status()
return response.text
except requests.RequestException:
return None
def submit_solution(self, pid, code):
files = {'code': ('solution.cpp', code)}
data = {'pid': pid, 'lang': 'cpp'}
try:
response = self.session.post(f"{self.judge_url}/submit", files=files, data=data)
response.raise_for_status()
return response.json().get('sid')
except requests.RequestException:
return None
def get_result(self, sid, poll_interval=60):
# Polls indefinitely until a result is returned
while True:
try:
response = self.session.get(f"{self.judge_url}/result/{sid}")
if response.status_code == 404:
time.sleep(poll_interval)
continue
response.raise_for_status()
result = response.json()
if result.get('status') in ['done', 'error']:
return result
time.sleep(poll_interval)
except requests.RequestException:
time.sleep(poll_interval)
def extract_cpp_code(response_text):
match = re.search(r'```cpp\n(.*?)```', response_text, re.DOTALL)
if match:
return match.group(1).strip()
return response_text
def get_llm_instance(model_name):
"""Factory function to instantiate the correct LLM class."""
if model_name == 'gemini':
return Gemini()
elif model_name == 'gpt':
return GPT()
elif model_name == 'claude':
return Claude()
elif model_name == 'claude-opus':
return Claude_Opus()
elif model_name == 'claude-sonnet-4-5':
return Claude_Sonnet_4_5()
elif model_name == 'gemini3':
return Gemini3()
elif model_name == 'Grok':
return Grok()
elif model_name == 'claude-opus-4-5':
return Claude_Opus_4_5()
else:
raise ValueError(f"Unknown model: {model_name}")
def process_single_attempt(pid, model_name):
"""
Generates one solution, judges it, and saves the results.
Returns the score (float/int) for the attempt.
Returns 0 if a critical error occurs before grading.
"""
run_id = f"{pid}_{model_name}"
result_filename = f"{solution_dir}/{run_id}_result.json"
solution_filename = f"{solution_dir}/{run_id}_solution.cpp"
print(f"[Processing {pid}] Starting attempt using {model_name}")
judge = LocalJudge()
try:
llm = get_llm_instance(model_name)
except ValueError as e:
print(e)
return 0
try:
# 1. Get Statement
statement = judge.get_problem_statement(pid)
if not statement:
print(f"[Processing {pid}] Failed to get statement.")
return 0
# 2. Generate Solution
llm_response, _ = llm.generate_solution(statement)
if not llm_response:
print(f"[Processing {pid}] LLM failed.")
final_result = {"status": "error", "error": "LLM_TIMEOUT_OR_FAILURE", "score": 0}
with open(result_filename, 'w', encoding='utf-8') as f:
json.dump(final_result, f, indent=4)
return 0
llm_text_content = str(llm_response)
solution_code = extract_cpp_code(llm_text_content)
with open(solution_filename, 'w', encoding='utf-8') as f:
f.write(solution_code)
# 3. Submit and Grade
with judge_throttler:
submission_id = judge.submit_solution(pid, solution_code)
if not submission_id:
final_result = {"status": "error", "error": "SUBMISSION_FAILED", "score": 0}
with open(result_filename, 'w', encoding='utf-8') as f:
json.dump(final_result, f, indent=4)
return 0
final_result = judge.get_result(submission_id)
# 4. Save Result
new_score = final_result.get('score', -1)
with open(result_filename, 'w', encoding='utf-8') as f:
json.dump(final_result, f, indent=4)
print(f"[Processing {pid}] Finished. Score: {new_score}")
return new_score
except Exception as e:
print(f"[Processing {pid}] A critical error occurred: {e}")
error_result = {"status": "error", "error": str(e), "score": 0}
try:
with open(result_filename, 'w', encoding='utf-8') as f:
json.dump(error_result, f, indent=4)
except Exception as e_file:
print(f" ...Additionally, FAILED to write error log: {e_file}")
return 0
if __name__ == "__main__":
# --- Argument Parsing ---
parser = argparse.ArgumentParser(description="Run LLM Judge Benchmarks")
parser.add_argument(
"model",
type=str,
help=f"The model to run. Options: {', '.join(AVAILABLE_MODELS)}"
)
args = parser.parse_args()
selected_model = args.model
if selected_model not in AVAILABLE_MODELS:
print(f"Error: '{selected_model}' is not a valid model.")
print(f"Available options: {', '.join(AVAILABLE_MODELS)}")
sys.exit(1)
if not os.path.exists(solution_dir):
os.makedirs(solution_dir)
judge = LocalJudge()
print("Fetching list of problems...")
problem_ids = judge.get_all_problems()
if not problem_ids:
print("No problems found on the judge.")
sys.exit(0)
print(f"Found {len(problem_ids)} problems. Starting single run for each using model: {selected_model}")
collected_scores = []
with concurrent.futures.ThreadPoolExecutor(max_workers=API_WORKERS) as executor:
futures = [executor.submit(process_single_attempt, pid, selected_model) for pid in problem_ids]
for future in concurrent.futures.as_completed(futures):
try:
# Retrieve the return value from the function (the score)
raw_score = future.result()
# Treat -1 (error/missing) as 0 for the average
if raw_score == -1:
effective_score = 0
else:
effective_score = raw_score
collected_scores.append(effective_score)
except Exception as e:
print(f"A job failed in the thread pool: {e}")
collected_scores.append(0) # Treat exceptions as 0 score
print("\nAll problems have been processed.")
# Calculate Average
if collected_scores:
avg_score = sum(collected_scores) / len(collected_scores)
print("-" * 30)
print(f"Final Average Score: {avg_score:.2f}")
print("-" * 30)
else:
print("No scores collected.") |