""" 人工验证界面 使用Gradio创建Web界面，用于人工验证生成的AIME题目 """ import json import os from typing import List, Dict, Any, Tuple from datetime import datetime import gradio as gr class HumanVerificationUI: """人工验证界面""" def __init__(self, data_path: str): """ 初始化验证界面 Args: data_path: 生成数据的JSON文件路径 """ self.data_path = data_path self.problems = self._load_problems() self.current_index = 0 self.verifications = self._load_verifications() def _load_problems(self) -> List[Dict[str, Any]]: """加载题目数据""" if not os.path.exists(self.data_path): raise FileNotFoundError(f"数据文件不存在: {self.data_path}") with open(self.data_path, 'r', encoding='utf-8') as f: return json.load(f) def _load_verifications(self) -> Dict[str, Any]: """加载已有的验证结果""" verification_path = self.data_path.replace(".json", "_verifications.json") if os.path.exists(verification_path): with open(verification_path, 'r', encoding='utf-8') as f: return json.load(f) return {} def _save_verifications(self): """保存验证结果""" verification_path = self.data_path.replace(".json", "_verifications.json") with open(verification_path, 'w', encoding='utf-8') as f: json.dump(self.verifications, f, ensure_ascii=False, indent=2) def get_current_problem(self) -> Tuple[str, str, str, str, str, str]: """获取当前题目信息""" if not self.problems: return "无题目", "", "", "", "", "0/0" problem = self.problems[self.current_index] problem_id = problem.get("id", "unknown") # 获取已有的验证信息 verification = self.verifications.get(problem_id, {}) return ( f"题目 {self.current_index + 1}/{len(self.problems)}", problem.get("problem", ""), f"答案: {problem.get('answer', 'N/A')}", problem.get("solution", ""), f"主题: {problem.get('topic', 'N/A')}", verification.get("comments", "") ) def verify_problem( self, correctness: int, clarity: int, difficulty_match: int, completeness: int, status: str, comments: str ) -> str: """ 验证当前题目 Args: correctness: 正确性评分 (1-5) clarity: 清晰度评分 (1-5) difficulty_match: 难度匹配评分 (1-5) completeness: 完整性评分 (1-5) status: 验证状态 (approved/rejected/needs_revision) comments: 评论 Returns: 验证结果消息 """ if not self.problems: return "❌ 无题目可验证" problem = self.problems[self.current_index] problem_id = problem.get("id", "unknown") # 保存验证结果 self.verifications[problem_id] = { "problem_id": problem_id, "scores": { "correctness": correctness, "clarity": clarity, "difficulty_match": difficulty_match, "completeness": completeness }, "total_score": (correctness + clarity + difficulty_match + completeness) / 4, "status": status, "comments": comments, "verified_at": datetime.now().isoformat() } self._save_verifications() return f"✅ 题目 {problem_id} 验证完成！\n总分: {self.verifications[problem_id]['total_score']:.2f}/5.0" def next_problem(self) -> Tuple[str, str, str, str, str, str]: """下一个题目""" if self.current_index < len(self.problems) - 1: self.current_index += 1 return self.get_current_problem() def prev_problem(self) -> Tuple[str, str, str, str, str, str]: """上一个题目""" if self.current_index > 0: self.current_index -= 1 return self.get_current_problem() def get_statistics(self) -> str: """获取验证统计信息""" if not self.verifications: return "暂无验证数据" total = len(self.problems) verified = len(self.verifications) approved = sum(1 for v in self.verifications.values() if v["status"] == "approved") rejected = sum(1 for v in self.verifications.values() if v["status"] == "rejected") needs_revision = sum(1 for v in self.verifications.values() if v["status"] == "needs_revision") avg_score = sum(v["total_score"] for v in self.verifications.values()) / verified if verified > 0 else 0 return f""" 📊 验证统计 总题目数: {total} 已验证: {verified} ({verified/total*100:.1f}%) 未验证: {total - verified} 验证结果: - ✅ 通过: {approved} - ❌ 拒绝: {rejected} - 🔄 需修改: {needs_revision} 平均评分: {avg_score:.2f}/5.0 """ def launch(self, share: bool = False): """启动Gradio界面""" with gr.Blocks(title="AIME题目人工验证") as demo: gr.Markdown("# 🎯 AIME题目人工验证系统") gr.Markdown(f"数据文件: `{self.data_path}`") with gr.Row(): with gr.Column(scale=2): # 题目显示区域 title = gr.Textbox(label="当前题目", interactive=False) problem_text = gr.Textbox(label="问题描述", lines=5, interactive=False) answer_text = gr.Textbox(label="答案", interactive=False) solution_text = gr.Textbox(label="解答过程", lines=10, interactive=False) metadata_text = gr.Textbox(label="元数据", interactive=False) with gr.Column(scale=1): # 评分区域 gr.Markdown("### 📝 评分 (1-5分)") correctness_slider = gr.Slider(1, 5, value=3, step=1, label="正确性") clarity_slider = gr.Slider(1, 5, value=3, step=1, label="清晰度") difficulty_slider = gr.Slider(1, 5, value=3, step=1, label="难度匹配") completeness_slider = gr.Slider(1, 5, value=3, step=1, label="完整性") # 状态选择 gr.Markdown("### ✅ 验证状态") status_radio = gr.Radio( choices=["approved", "rejected", "needs_revision"], value="approved", label="状态" ) # 评论 comments_text = gr.Textbox(label="评论", lines=3, placeholder="请输入评论...") # 验证按钮 verify_btn = gr.Button("✅ 提交验证", variant="primary") verify_result = gr.Textbox(label="验证结果", interactive=False) # 导航按钮 with gr.Row(): prev_btn = gr.Button("⬅️ 上一题") next_btn = gr.Button("下一题 ➡️") # 统计信息 with gr.Row(): stats_text = gr.Textbox(label="验证统计", lines=10, interactive=False) refresh_stats_btn = gr.Button("🔄 刷新统计") # 加载初始题目 demo.load( fn=self.get_current_problem, outputs=[title, problem_text, answer_text, solution_text, metadata_text, comments_text] ) # 绑定事件 verify_btn.click( fn=self.verify_problem, inputs=[correctness_slider, clarity_slider, difficulty_slider, completeness_slider, status_radio, comments_text], outputs=verify_result ) next_btn.click( fn=self.next_problem, outputs=[title, problem_text, answer_text, solution_text, metadata_text, comments_text] ) prev_btn.click( fn=self.prev_problem, outputs=[title, problem_text, answer_text, solution_text, metadata_text, comments_text] ) refresh_stats_btn.click( fn=self.get_statistics, outputs=stats_text ) demo.launch(share=share, server_name="127.0.0.1", server_port=7860) if __name__ == "__main__": import sys if len(sys.argv) < 2: print("用法: python human_verification_ui.py ") print("示例: python human_verification_ui.py generated_data/aime_generated_20250110_120000.json") sys.exit(1) data_path = sys.argv[1] ui = HumanVerificationUI(data_path) ui.launch(share=False)