feat(val-blog): add 2026-04-30 dream journey post

2026-08-13 17:00:22 +08:00
parent 691f96e8a6
commit 81978704fc
941 changed files with 195468 additions and 755 deletions
@@ -0,0 +1,39 @@
+#!/bin/bash
+# download-model.sh - 下载适合 8GB 内存的 Qwen 模型
+
+MODEL_DIR="${HOME}/.llama/models"
+mkdir -p "$MODEL_DIR"
+
+echo "=== 下载 Qwen 模型 (适合 8GB 内存) ==="
+echo ""
+echo "可选模型:"
+echo "1) Qwen3-1.8B-Q4_K_M (推荐, ~1.2GB, 速度最快)"
+echo "2) Qwen3-4B-Q4_K_M (~2.5GB, 质量更好)"
+echo "3) Qwen2.5-1.5B-Q4_K_M (备用, ~1GB)"
+echo ""
+
+# 默认下载 1.8B 版本
+MODEL_URL="https://huggingface.co/Qwen/Qwen3-1.8B-GGUF/resolve/main/qwen3-1.8b-q4_k_m.gguf"
+MODEL_NAME="qwen3-1.8b-q4_k_m.gguf"
+
+cd "$MODEL_DIR"
+
+if [ -f "$MODEL_NAME" ]; then
+    echo "✅ 模型已存在: $MODEL_NAME"
+else
+    echo "⬇️ 正在下载 $MODEL_NAME..."
+    echo "(如下载慢可用浏览器访问 https://huggingface.co/Qwen/Qwen3-1.8B-GGUF)"
+    
+    # 使用 curl 下载，支持断点续传
+    curl -L -C - -o "$MODEL_NAME" "$MODEL_URL" || {
+        echo "❌ 下载失败，请手动下载到: $MODEL_DIR/$MODEL_NAME"
+        echo "下载链接: $MODEL_URL"
+        exit 1
+    }
+fi
+
+echo ""
+echo "✅ 模型已准备: $MODEL_DIR/$MODEL_NAME"
+echo ""
+echo "=== 下一步 ==="
+echo "启动服务: ./start-server.sh"
@@ -0,0 +1,26 @@
+#!/bin/bash
+# install-llama-cpp.sh - 在 Intel Mac 上安装优化版 llama.cpp
+
+set -e
+
+echo "=== 安装 llama.cpp (Intel Mac 优化版) ==="
+
+# 方式1: 通过 Homebrew 安装（最简单）
+if command -v brew &> /dev/null; then
+    echo "通过 Homebrew 安装..."
+    brew install llama.cpp
+    echo "✅ Homebrew 安装完成"
+else
+    echo "⚠️ 未检测到 Homebrew，建议先安装: https://brew.sh"
+    echo "或使用方式2手动编译"
+fi
+
+echo ""
+echo "=== 安装后验证 ==="
+which llama-cli || echo "llama-cli 未找到"
+which llama-server || echo "llama-server 未找到"
+
+echo ""
+echo "=== 下一步 ==="
+echo "1. 下载模型: 运行 download-model.sh"
+echo "2. 启动服务: 运行 start-server.sh"
@@ -0,0 +1,263 @@
+#!/usr/bin/env python3
+"""
+FTS5 记忆索引系统
+用于全文搜索 memory/*.md 文件
+"""
+
+import sqlite3
+import os
+import re
+from pathlib import Path
+from datetime import datetime
+
+# 配置
+MEMORY_DIR = Path.home() / ".openclaw" / "workspace" / "memory"
+DB_PATH = MEMORY_DIR / ".fts5" / "memory.db"
+
+
+def init_db():
+    """初始化 FTS5 数据库"""
+    DB_PATH.parent.mkdir(parents=True, exist_ok=True)
+    
+    conn = sqlite3.connect(DB_PATH)
+    cursor = conn.cursor()
+    
+    # 创建文件表
+    cursor.execute("""
+        CREATE TABLE IF NOT EXISTS files (
+            id INTEGER PRIMARY KEY,
+            path TEXT UNIQUE NOT NULL,
+            modified_at TIMESTAMP,
+            indexed_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+        )
+    """)
+    
+    # 创建 FTS5 虚拟表用于全文搜索
+    cursor.execute("""
+        CREATE VIRTUAL TABLE IF NOT EXISTS memory_index USING fts5(
+            content,
+            file_id,
+            chunk_index,
+            tokenize='porter unicode61'
+        )
+    """)
+    
+    # 创建 chunks 表存储分块信息
+    cursor.execute("""
+        CREATE TABLE IF NOT EXISTS chunks (
+            id INTEGER PRIMARY KEY,
+            file_id INTEGER,
+            chunk_index INTEGER,
+            content TEXT,
+            start_line INTEGER,
+            end_line INTEGER,
+            FOREIGN KEY (file_id) REFERENCES files(id)
+        )
+    """)
+    
+    conn.commit()
+    conn.close()
+    print(f"✓ 数据库初始化完成: {DB_PATH}")
+
+
+def chunk_content(content, chunk_size=500, overlap=100):
+    """
+    将内容分块，带重叠以保持上下文
+    
+    Args:
+        content: 文件内容
+        chunk_size: 每块字符数
+        overlap: 重叠字符数
+    
+    Returns:
+        list of (chunk_text, start_line, end_line)
+    """
+    lines = content.split('\n')
+    chunks = []
+    
+    current_chunk = []
+    current_size = 0
+    start_line = 0
+    
+    for i, line in enumerate(lines):
+        line_with_newline = line + '\n'
+        
+        if current_size + len(line_with_newline) > chunk_size and current_chunk:
+            # 保存当前块
+            chunk_text = ''.join(current_chunk)
+            chunks.append((chunk_text, start_line + 1, i))
+            
+            # 重叠部分
+            overlap_text = chunk_text[-overlap:] if len(chunk_text) > overlap else chunk_text
+            current_chunk = [overlap_text]
+            current_size = len(overlap_text)
+            start_line = i - len(overlap_text.split('\n')) + 1
+        
+        current_chunk.append(line_with_newline)
+        current_size += len(line_with_newline)
+    
+    # 最后一块
+    if current_chunk:
+        chunk_text = ''.join(current_chunk)
+        chunks.append((chunk_text, start_line + 1, len(lines)))
+    
+    return chunks
+
+
+def index_file(file_path, conn=None):
+    """索引单个文件"""
+    close_conn = False
+    if conn is None:
+        conn = sqlite3.connect(DB_PATH)
+        close_conn = True
+    
+    cursor = conn.cursor()
+    
+    # 读取文件
+    try:
+        with open(file_path, 'r', encoding='utf-8') as f:
+            content = f.read()
+    except Exception as e:
+        print(f"✗ 读取失败 {file_path}: {e}")
+        return
+    
+    # 获取文件信息
+    stat = os.stat(file_path)
+    modified_at = datetime.fromtimestamp(stat.st_mtime)
+    rel_path = str(Path(file_path).relative_to(MEMORY_DIR.parent))
+    
+    # 插入或更新文件记录
+    cursor.execute("""
+        INSERT OR REPLACE INTO files (path, modified_at, indexed_at)
+        VALUES (?, ?, CURRENT_TIMESTAMP)
+    """, (rel_path, modified_at))
+    
+    file_id = cursor.lastrowid
+    
+    # 删除旧索引
+    cursor.execute("DELETE FROM memory_index WHERE file_id = ?", (file_id,))
+    cursor.execute("DELETE FROM chunks WHERE file_id = ?", (file_id,))
+    
+    # 分块并索引
+    chunks = chunk_content(content)
+    
+    for chunk_idx, (chunk_text, start_line, end_line) in enumerate(chunks):
+        # 插入 chunks 表
+        cursor.execute("""
+            INSERT INTO chunks (file_id, chunk_index, content, start_line, end_line)
+            VALUES (?, ?, ?, ?, ?)
+        """, (file_id, chunk_idx, chunk_text, start_line, end_line))
+        
+        # 插入 FTS5 索引
+        cursor.execute("""
+            INSERT INTO memory_index (content, file_id, chunk_index)
+            VALUES (?, ?, ?)
+        """, (chunk_text, file_id, chunk_idx))
+    
+    conn.commit()
+    
+    if close_conn:
+        conn.close()
+    
+    print(f"✓ 已索引: {rel_path} ({len(chunks)} 块)")
+
+
+def index_all():
+    """索引所有 memory 文件"""
+    init_db()
+    
+    conn = sqlite3.connect(DB_PATH)
+    
+    # 查找所有 .md 文件
+    md_files = list(MEMORY_DIR.glob("*.md"))
+    
+    print(f"\n开始索引 {len(md_files)} 个文件...")
+    
+    for file_path in md_files:
+        if file_path.name.startswith('.'):
+            continue
+        index_file(file_path, conn)
+    
+    conn.close()
+    print(f"\n✓ 索引完成")
+
+
+def search(query, limit=10):
+    """
+    全文搜索
+    
+    Args:
+        query: 搜索关键词
+        limit: 返回结果数量
+    
+    Returns:
+        list of dict with path, snippet, rank
+    """
+    conn = sqlite3.connect(DB_PATH)
+    cursor = conn.cursor()
+    
+    # 使用 FTS5 的 snippet 功能生成摘要
+    cursor.execute("""
+        SELECT 
+            f.path,
+            snippet(memory_index, 0, '[', ']', '...', 32) as snippet,
+            c.start_line,
+            c.end_line,
+            rank
+        FROM memory_index
+        JOIN files f ON memory_index.file_id = f.id
+        JOIN chunks c ON memory_index.file_id = c.file_id 
+                     AND memory_index.chunk_index = c.chunk_index
+        WHERE memory_index MATCH ?
+        ORDER BY rank
+        LIMIT ?
+    """, (query, limit))
+    
+    results = []
+    for row in cursor.fetchall():
+        results.append({
+            'path': row[0],
+            'snippet': row[1],
+            'start_line': row[2],
+            'end_line': row[3],
+            'rank': row[4]
+        })
+    
+    conn.close()
+    return results
+
+
+def print_search_results(query, results):
+    """打印搜索结果"""
+    print(f"\n🔍 搜索: '{query}'")
+    print(f"找到 {len(results)} 条结果:\n")
+    
+    for i, r in enumerate(results, 1):
+        print(f"{i}. {r['path']} (行 {r['start_line']}-{r['end_line']})")
+        print(f"   {r['snippet']}")
+        print()
+
+
+if __name__ == "__main__":
+    import sys
+    
+    if len(sys.argv) < 2:
+        print("用法:")
+        print(f"  {sys.argv[0]} index       # 索引所有文件")
+        print(f"  {sys.argv[0]} search <query>  # 搜索")
+        sys.exit(1)
+    
+    command = sys.argv[1]
+    
+    if command == "index":
+        index_all()
+    elif command == "search":
+        if len(sys.argv) < 3:
+            print("错误: 需要提供搜索关键词")
+            sys.exit(1)
+        query = sys.argv[2]
+        results = search(query)
+        print_search_results(query, results)
+    else:
+        print(f"未知命令: {command}")
+        sys.exit(1)
@@ -0,0 +1,33 @@
+# OpenClaw 本地模型集成配置
+# 将此配置添加到 ~/.openclaw/openclaw.json 的 models.providers 部分
+
+{
+  "local": {
+    "baseUrl": "http://127.0.0.1:8080/v1",
+    "apiKey": "local-llm-key",
+    "api": "openai-completions",
+    "models": [
+      {
+        "id": "qwen3-1.8b",
+        "name": "Qwen3-1.8B (Local)",
+        "reasoning": false,
+        "input": ["text"],
+        "cost": {
+          "input": 0,
+          "output": 0,
+          "cacheRead": 0,
+          "cacheWrite": 0
+        },
+        "contextWindow": 4096,
+        "maxTokens": 2048
+      }
+    ]
+  }
+}
+
+# 然后修改 agents.defaults.model.fallbacks 将本地模型加入列表
+# "fallbacks": [
+#   "lkeap/kimi-k2.5",
+#   "local/qwen3-1.8b",    <-- 添加这行
+#   ...
+# ]
@@ -0,0 +1,51 @@
+#!/bin/bash
+# start-server.sh - 启动 llama.cpp 推理服务 (Intel Mac 8GB 内存优化)
+
+MODEL_DIR="${HOME}/.llama/models"
+MODEL_NAME="qwen3-1.8b-q4_k_m.gguf"
+MODEL_PATH="$MODEL_DIR/$MODEL_NAME"
+
+# 检查模型是否存在
+if [ ! -f "$MODEL_PATH" ]; then
+    echo "❌ 模型未找到: $MODEL_PATH"
+    echo "请先运行: ./download-model.sh"
+    exit 1
+fi
+
+echo "=== 启动 llama.cpp 推理服务 ==="
+echo "模型: $MODEL_NAME"
+echo "内存优化: 针对 Intel Mac 8GB 优化"
+echo ""
+
+# Intel CPU + 8GB 内存优化参数
+# -t 4: 使用4线程 (你的CPU是4核8线程，留一些给系统)
+# -c 4096: 上下文长度 4K (省内存)
+# -np 1: 单并发 (避免内存爆炸)
+# --mlock: 锁定内存防止 swap (可选，如果你的内存够)
+# --host 0.0.0.0: 允许局域网访问 (可选)
+
+llama-server \
+    --model "$MODEL_PATH" \
+    --threads 4 \
+    --ctx-size 4096 \
+    --parallel 1 \
+    --batch-size 512 \
+    --timeout 300 \
+    --host 127.0.0.1 \
+    --port 8080 \
+    --api-key "local-llm-key" \
+    --verbose \
+    "$@"
+
+# 如果 llama-server 找不到，尝试 llama-cli
+if [ $? -ne 0 ]; then
+    echo ""
+    echo "⚠️ llama-server 未找到，尝试 llama-cli 交互模式..."
+    echo "(注意: llama-cli 不提供 API 服务，仅用于测试)"
+    llama-cli \
+        --model "$MODEL_PATH" \
+        --threads 4 \
+        --ctx-size 4096 \
+        --interactive \
+        --verbose-prompt
+fi