Spaces:
Sleeping
Sleeping
File size: 10,746 Bytes
7a9b69c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 |
import gradio as gr
import requests
import random
from datasets import load_dataset, Dataset
from typing import Dict, List
import re
import datetime
import pandas as pd
import os
from dotenv import load_dotenv
load_dotenv()
def sanitize_theme_name(theme: str) -> str:
sanitized = re.sub(r'[^\w\s-]', '', theme)
sanitized = re.sub(r'[-\s]+', '_', sanitized)
return sanitized.lower().strip('_')
def load_questions_from_dataset() -> Dict[str, List[Dict]]:
dataset = load_dataset("SASLeaderboard/sas_opposition_exam_data")
dataset = dataset['train'].filter(lambda x: x['theme'] == 'FEA UrologΓa')
questions_by_theme = {}
skipped = 0
loaded = 0
for item in dataset:
theme = item['theme']
answers = item.get('answers', [])
correct_answer = item.get('correct_answer', '')
if not answers or not correct_answer or len(answers) < 3:
skipped += 1
continue
while len(answers) < 4:
answers.append(answers[-1])
sanitized_theme = sanitize_theme_name(theme)
if sanitized_theme not in questions_by_theme:
questions_by_theme[sanitized_theme] = []
try:
question = {
"statement": item['statement'],
"options": {
"A": answers[0],
"B": answers[1],
"C": answers[2],
"D": answers[3]
},
"real_answer": correct_answer,
"theme": theme,
"sanitized_theme": sanitized_theme,
"version": item.get('version', 'Default')
}
questions_by_theme[sanitized_theme].append(question)
loaded += 1
except Exception as e:
skipped += 1
continue
print(f"Loaded {loaded} questions, skipped {skipped} invalid questions")
return questions_by_theme
def ask_ai_model(api_key: str, model: str, question: Dict) -> tuple:
prompt = f"""You are a medical expert taking a urology examination. Please analyze this question carefully and provide your answer.
Question: {question['statement']}
Options:
A) {question['options']['A']}
B) {question['options']['B']}
C) {question['options']['C']}
D) {question['options']['D']}
Please provide your answer in this exact format:
Answer: [A/B/C/D]
Then provide your reasoning."""
try:
headers = {
"Authorization": f"Bearer {api_key}",
"Content-Type": "application/json"
}
data = {
"model": model,
"messages": [
{"role": "user", "content": prompt}
]
}
response = requests.post("https://openrouter.ai/api/v1/chat/completions",
headers=headers, json=data)
if response.status_code == 200:
result = response.json()
ai_response = result["choices"][0]["message"]["content"]
ai_answer = extract_answer_from_response(ai_response)
return ai_response, ai_answer
else:
error_msg = f"API Error {response.status_code}: {response.text}"
return error_msg, "API_ERROR"
except Exception as e:
error_msg = f"Request Error: {str(e)}"
return error_msg, "REQUEST_ERROR"
def extract_answer_from_response(ai_response: str) -> str:
if not ai_response:
return "EMPTY_RESPONSE"
lines = ai_response.split('\n')
for line in lines:
line_clean = line.strip().lower()
if line_clean.startswith('answer:'):
answer_part = line.split(':')[1].strip().upper()
for char in answer_part:
if char in ['A', 'B', 'C', 'D']:
return char
for line in lines:
line_clean = line.strip().lower()
if 'answer is' in line_clean:
for char in ['A', 'B', 'C', 'D']:
if char.lower() in line_clean.split('answer is')[1][:5]:
return char
for line in lines[:5]:
line_upper = line.upper()
for char in ['A', 'B', 'C', 'D']:
patterns = [f"{char})", f"{char}.", f"OPTION {char}", f"({char})", f"CHOICE {char}"]
for pattern in patterns:
if pattern in line_upper:
return char
for line in lines[:3]:
for char in ['A', 'B', 'C', 'D']:
if char in line.upper():
return char
for char in ['A', 'B', 'C', 'D']:
if char in ai_response.upper():
return char
return "NO_ANSWER_FOUND"
def save_results_to_dataset(results: List[Dict], hf_token: str = None) -> str:
if not results:
return "No results to save"
if not hf_token:
hf_token = os.getenv("HF_TOKEN")
if not hf_token:
return "β HuggingFace token not found. Please provide it in the interface or set HF_TOKEN environment variable"
try:
try:
existing_dataset = load_dataset("SASLeaderboard/results", use_auth_token=hf_token)
existing_data = existing_dataset['train'].to_pandas()
except Exception:
existing_data = None
new_data = pd.DataFrame(results)
if existing_data is not None:
combined_data = pd.concat([existing_data, new_data], ignore_index=True)
else:
combined_data = new_data
new_dataset = Dataset.from_pandas(combined_data)
new_dataset.push_to_hub(
"SASLeaderboard/results",
token=hf_token,
commit_message=f"Automated exam results for {results[0]['model']} - {len(results)} questions"
)
return f"β
Successfully saved {len(results)} results to SASLeaderboard/results dataset"
except Exception as e:
return f"β Error saving results: {str(e)}"
def run_automated_exam(api_key: str, model: str, hf_token: str = ""):
if not api_key:
yield "β Please provide OpenRouter API key"
return
if not model:
yield "β Please provide model name"
return
yield "π Loading questions from dataset..."
try:
all_questions_by_theme = load_questions_from_dataset()
all_questions = []
for theme_questions in all_questions_by_theme.values():
all_questions.extend(theme_questions)
total_questions = len(all_questions)
yield f"β
Loaded {total_questions} questions from dataset"
yield f"π Starting automated exam with ALL {total_questions} questions for model: {model}"
session_id = f"{model}_{datetime.datetime.now().strftime('%Y%m%d_%H%M%S')}"
results = []
correct_count = 0
for i, question in enumerate(all_questions):
ai_response, ai_answer = ask_ai_model(api_key, model, question)
if ai_answer in ["API_ERROR", "REQUEST_ERROR", "EMPTY_RESPONSE", "NO_ANSWER_FOUND"]:
yield f"β οΈ Question {i+1}: Error getting answer - {ai_answer}. Response: {ai_response[:100]}..."
is_correct = ai_answer == question['real_answer']
if is_correct:
correct_count += 1
result = {
"session_id": session_id,
"model": model,
"question": question['statement'],
"theme": question['theme'],
"correct_answer": question['real_answer'],
"ai_answer": ai_answer,
"ai_response": ai_response,
"is_correct": is_correct,
"timestamp": datetime.datetime.now().isoformat(),
"options_a": question['options']['A'],
"options_b": question['options']['B'],
"options_c": question['options']['C'],
"options_d": question['options']['D']
}
results.append(result)
current_accuracy = (correct_count / (i + 1)) * 100
status_emoji = "β
" if is_correct else "β"
yield f"{status_emoji} Q{i+1}/{total_questions}: Accuracy: {correct_count}/{i+1} ({current_accuracy:.1f}%) | AI: {ai_answer} vs Correct: {question['real_answer']} | {question['statement'][:80]}..."
yield f"πΎ Saving results to HuggingFace dataset..."
save_result = save_results_to_dataset(results, hf_token)
final_accuracy = (correct_count / len(results)) * 100
yield f"""
## π― Exam Complete!
**Final Results:**
- Model: {model}
- Total Questions: {len(results)}
- Correct Answers: {correct_count}
- Final Accuracy: {final_accuracy:.1f}%
- Session ID: {session_id}
**Save Status:** {save_result}
The automated exam has been completed successfully!
"""
except Exception as e:
yield f"β Error during automated exam: {str(e)}"
with gr.Blocks(title="Automated Urology Exam System") as demo:
gr.Markdown("# Automated Urology Exam System")
gr.Markdown("This system automatically runs a complete urology exam for AI models using ALL available questions (~150) and saves results to the dataset.")
with gr.Row():
with gr.Column():
gr.Markdown("**Get your API key:** [OpenRouter Keys](https://openrouter.ai/settings/keys)")
api_key_input = gr.Textbox(
label="OpenRouter API Key",
type="password",
placeholder="Enter your OpenRouter API key"
)
with gr.Column():
gr.Markdown("**Find models:** [OpenRouter Models](https://openrouter.ai/models)")
model_input = gr.Textbox(
label="Model Name",
placeholder="e.g., anthropic/claude-3-sonnet",
value="anthropic/claude-3-sonnet"
)
with gr.Row():
start_exam_btn = gr.Button("Start Automated Exam", variant="primary", size="lg")
with gr.Row():
progress_output = gr.Textbox(
label="Exam Progress - Dont close this window",
placeholder="Exam progress will be displayed here...",
lines=15,
max_lines=20,
interactive=False
)
start_exam_btn.click(
run_automated_exam,
inputs=[api_key_input, model_input],
outputs=[progress_output]
)
if __name__ == "__main__":
demo.launch() |