chore(vault): update Education AI research memory and tools

2026-07-06 15:17:40 +08:00
parent b101b6516c
commit 54a823a6ea
53 changed files with 46913 additions and 28 deletions
@@ -0,0 +1,196 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+使用 singlestep 提示词对测试文件生成 WIKI
+直接使用当前 opencode 配置的 LLM
+"""
+
+import os
+import re
+import json
+from datetime import datetime
+import openai
+
+# 配置
+base_dir = os.getcwd()
+experiment_dir = os.path.join(base_dir, "tools", "experiments", "wiki-generation-compare")
+source_file = os.path.join(base_dir, "raw", "呼吸之间_李谨伯", "第一编 从身体入手.md")
+output_dir = os.path.join(experiment_dir, "output", "group-b-real")
+prompt_file = os.path.join(experiment_dir, "prompts", "singlestep.md")
+
+# 创建输出目录
+os.makedirs(output_dir, exist_ok=True)
+os.makedirs(os.path.join(output_dir, "wiki", "concepts"), exist_ok=True)
+os.makedirs(os.path.join(output_dir, "wiki", "methods"), exist_ok=True)
+os.makedirs(os.path.join(output_dir, "wiki", "entities"), exist_ok=True)
+
+# 记录开始时间
+start_time = datetime.now()
+start_time_str = start_time.strftime("%Y-%m-%d %H:%M:%S")
+
+print("=== B 组测试：Single-Step 模式（使用 9router reasoning 模型）===")
+print(f"开始时间: {start_time_str}")
+
+# 读取源文件
+print(f"\n读取源文件: {source_file}")
+with open(source_file, 'r', encoding='utf-8') as f:
+    source_content = f.read()
+print(f"  源文件长度: {len(source_content)} 字符")
+
+# 读取 prompt
+print(f"\n读取 prompt: {prompt_file}")
+with open(prompt_file, 'r', encoding='utf-8') as f:
+    template = f.read()
+
+# 替换 {SOURCE_CONTENT}
+full_prompt = template.replace("{SOURCE_CONTENT}", source_content)
+print(f"  Prompt 长度: {len(full_prompt)} 字符")
+
+# 检查 API Key
+api_key = os.environ.get('OPENAI_API_KEY')
+if not api_key:
+    print("[ERROR] 未找到 OPENAI_API_KEY 环境变量")
+    exit(1)
+print(f"  API Key 长度: {len(api_key)}")
+
+# 尝试不同的模型名称
+models_to_try = [
+    "9router/reasoning",
+    "reasoning",
+    "openai/gpt-4o-mini",
+    "gpt-4o-mini",
+    "openai/gpt-3.5-turbo",
+    "gpt-3.5-turbo"
+]
+
+# 尝试连接并获取可用模型
+print("\n查找可用模型...")
+available_models = []
+for model_name in models_to_try:
+    try:
+        client = openai.OpenAI(api_key=api_key, timeout=60)
+        # 先尝试简单的 chat completion
+        test_response = client.chat.completions.create(
+            model=model_name,
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=5
+        )
+        print(f"  [OK] 模型可用: {model_name}")
+        available_models.append(model_name)
+        break
+    except Exception as e:
+        print(f"  [X] {model_name}: {type(e).__name__}")
+        if "404" not in str(e) and "not exist" not in str(e):
+            # 如果不是 404，可能是网络问题
+            print(f"    错误详情: {str(e)[:200]}")
+
+if not available_models:
+    print("\n[ERROR] 没有可用的模型")
+    exit(1)
+
+# 使用第一个可用的模型
+selected_model = available_models[0]
+print(f"\n使用模型: {selected_model}")
+
+# 初始化客户端
+client = openai.OpenAI(api_key=api_key, timeout=300)
+
+# 调用 LLM 生成
+print("\n调用 LLM 生成 Wiki 页面...")
+try:
+    response = client.chat.completions.create(
+        model=selected_model,
+        messages=[
+            {
+                "role": "system",
+                "content": "你是一位知识库构建专家。请直接从源文件中提取关键术语并生成 Wiki 页面。"
+            },
+            {
+                "role": "user",
+                "content": full_prompt
+            }
+        ],
+        temperature=0.5,
+        max_tokens=16000
+    )
+
+    generation_result = response.choices[0].message.content
+    print(f"  [OK] LLM 生成完成")
+    print(f"  生成内容长度: {len(generation_result)} 字符")
+
+    # 保存原始生成结果
+    raw_output_file = os.path.join(output_dir, "raw_generation.md")
+    with open(raw_output_file, 'w', encoding='utf-8') as f:
+        f.write(generation_result)
+    print(f"  [OK] 原始生成结果已保存到: {raw_output_file}")
+
+except Exception as e:
+    print(f"  [ERROR] LLM API 调用失败: {type(e).__name__}: {e}")
+    exit(1)
+
+# 解析生成结果
+print("\n解析生成的 Wiki 页面...")
+sample_pages = []
+
+# 解析 ---FILE: ... ---END FILE--- 块
+file_blocks = re.findall(r'---FILE: (.*?)---(.*?)---END FILE---', generation_result, re.DOTALL)
+
+if not file_blocks:
+    print("  [WARNING] 未找到标准文件块格式，尝试宽松匹配...")
+    file_blocks = re.findall(r'FILE: (.*?)\n(.*?)(?=(?:FILE:|$))', generation_result, re.DOTALL)
+
+print(f"  解析到 {len(file_blocks)} 个文件块")
+
+for i, block in enumerate(file_blocks):
+    if isinstance(block, tuple):
+        file_path = block[0].strip()
+        file_content = block[1].strip()
+    else:
+        continue
+
+    # 创建目录（如果需要）
+    full_path = os.path.join(output_dir, file_path)
+    file_dir = os.path.dirname(full_path)
+    os.makedirs(file_dir, exist_ok=True)
+
+    # 写入文件
+    with open(full_path, 'w', encoding='utf-8') as f:
+        f.write(file_content)
+    sample_pages.append(full_path)
+    print(f"  [OK] 生成 [{i+1}/{len(file_blocks)}]: {file_path}")
+
+# 记录结束时间
+end_time = datetime.now()
+end_time_str = end_time.strftime("%Y-%m-%d %H:%M:%S")
+duration = (end_time - start_time).total_seconds() / 60
+
+print(f"\n=== B 组测试完成 ===")
+print(f"结束时间: {end_time_str}")
+print(f"总耗时: {round(duration, 2)} 分钟")
+print(f"生成文件数: {len(sample_pages)}")
+
+# 保存元数据
+metadata = {
+    "group": "B-real",
+    "mode": "Single-Step",
+    "sourceFile": "raw/呼吸之间_李谨伯/第一编 从身体入手.md",
+    "startTime": start_time_str,
+    "endTime": end_time_str,
+    "durationMinutes": round(duration, 2),
+    "pagesGenerated": len(sample_pages),
+    "model": selected_model,
+    "rawContentLength": len(generation_result),
+    "status": "success"
+}
+
+metadata_file = os.path.join(output_dir, "metadata.json")
+with open(metadata_file, 'w', encoding='utf-8') as f:
+    json.dump(metadata, f, ensure_ascii=False, indent=2)
+
+print(f"\n[OK] 元数据已保存到: {metadata_file}")
+print(f"\n输出文件位置:")
+print(f"  原始生成: {raw_output_file}")
+print(f"  Wiki 页面: {output_dir}/wiki/")
+print(f"  元数据: {metadata_file}")
+print(f"\n接下来可以运行评估脚本对比 A 组和 B 组的结果：")
+print(f"  python tools/experiments/wiki-generation-compare/scripts/evaluate-real.py")