惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Microsoft Security Blog
Microsoft Security Blog
J
Java Code Geeks
GbyAI
GbyAI
aimingoo的专栏
aimingoo的专栏
L
LangChain Blog
I
InfoQ
D
Docker
F
Fortinet All Blogs
Y
Y Combinator Blog
Martin Fowler
Martin Fowler
月光博客
月光博客
B
Blog
Engineering at Meta
Engineering at Meta
T
Tailwind CSS Blog
罗磊的独立博客
博客园_首页
G
Google Developers Blog
Stack Overflow Blog
Stack Overflow Blog
Recent Announcements
Recent Announcements
D
DataBreaches.Net
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
B
Blog RSS Feed
IT之家
IT之家
V
V2EX

Hacker News: Show HN

PurrrrrFocus: Pomodoro Timer App - App Store Workflow Engine — Multi-Step Orchestration for Bun RapidPhoto: Pro Photo Editor App - App Store GitHub - DheerG/swarms: Achieve extraordinary results with claude code across a variety of tasks SPICE simulation → oscilloscope → verification with Claude Code — Lucas Gerads Show HN: VCoding – A 5 MB native Windows IDE with no dynamic dependencies Show HN: LLMs don't hallucinate because they're bad at math, it's the format GitHub - Agent-FM/agentfm-core: AgentFM is a peer-to-peer network that turns everyday computers into a decentralized AI supercomputer. AgentFM lets you run massive AI workloads directly across a global mesh of idle CPUs and GPUs. Show HN: Tracking Top US Science Olympiad Alumni over Last 25 Years GitHub - Potarix/agent-hub: One place to talk to all your agents Show HN: Runtime security for AI agents(injection,tool abuse, data exfiltration) GitHub - dubeyKartikay/lazyspotify: Terminal Spotify client for macOS and Linux GitHub - the-banana-tool/king-louie: Easy to use GUI Personal AI Assistant. Win/Linux/Mac. Show HN I made my vacation rental bookable by AI agents–no Airbnb, 0% commission GitHub - basteez/jsf-autoreload: maven plugin to enable hot reload on jsf projects uvm32/hosts/host-gdbstub at main · ringtailsoftware/uvm32 GitHub - labsai/EDDI: Config-driven engine that turns JSON into production-grade AI agents. Multi-agent orchestration, 12+ LLM providers, MCP/A2A protocols, RAG, persistent memory, and enterprise compliance (EU AI Act, GDPR, HIPAA). Built on Quarkus. GitHub - glitchnsec/fortyone-oss: AI Executive Assistant Platform Quickstart | Alien GitHub - muxshed/shed: One stream in, or many. Every destination, simultaneously. No cloud middleman, no per-channel fees, no limits. GitHub - ocrbase-hq/ocrbase: 📄 PDF/IMG ->.MD/JSON Document OCR API for PaddleOCR and GLMOCR. Self-hostable. GitHub - impactjo/home-memory: MCP server that lets your AI assistant remember everything about your home. GitHub - Sets88/dbcls: DbCls is a powerful terminal database client that supports various databases GitHub - neptun2000/heor-agent-mcp GitHub - SeanFDZ/macmind: Single-layer transformer in HyperTalk for the classic Macintosh RollQuation: Math Puzzles - Apps on Google Play GitHub - dropbox/witchcraft Show HN: Agent-cache – Multi-tier LLM/tool/session caching for Valkey and Redis GitHub - opentalon/opentalon: OpenTalon is an open-source platform built from the ground up in Go as a robust alternative to OpenClaw LinkedIn™ 职位抓取工具 - Chrome 应用商店
GitHub - Rand01ph/gemma-trans: macOS 本地 AI 划词翻译:Ge...
Rand01ph · 2026-06-27 · via Hacker News: Show HN

GemmaTrans

macOS 本地大模型划词翻译。基于 MLX-Swift(Apple Silicon 原生加速),支持多个本地翻译模型(Google Gemma 4 / 腾讯混元 Hy-MT2),完全离线运行:

  • 本地 HTTP API127.0.0.1:8765):极简 /translate 接口 + 兼容 /v1/chat/completions 请求格式,PopClip、Bob、Raycast 等工具直连
  • menu bar app:全局热键划词翻译,浮窗流式显示译文
  • 智能双向:自动检测语言——中文 → 英文,其他语言 → 中文(目标语言可配置)
  • 多模型可选:内置 Gemma 4(E4B/E2B,4-bit)与腾讯混元 Hy-MT2(1.8B 翻译专用,4/8-bit);设置页可下载、切换、删除模型,支持后台下载(边用边下)
  • 性能可视:翻译后显示每秒 token 速率(按模型分别记录),随时观察各模型快慢
  • 模型下载:首次启动按内存自动选 Gemma 档拉取(E4B≈4.9GB / E2B≈3.6GB),双源(HuggingFace / 国内 ModelScope)+ 断点续传

快速开始

构建并启动 API 服务

git clone https://github.com/Rand01ph/gemma-trans && cd gemma-trans
# MLX 的 Metal 着色器需要 xcodebuild(首次如提示缺 Metal 工具链:xcodebuild -downloadComponent MetalToolchain)
xcodebuild -scheme gemma-trans-cli -destination 'platform=macOS' -skipMacroValidation -derivedDataPath .build-cli build
.build-cli/Build/Products/Debug/gemma-trans-cli serve
# 首次自动下载模型 → Model ready. Listening on http://127.0.0.1:8765

国内网络可用镜像:启动前 export HF_ENDPOINT=https://hf-mirror.comgemma-trans-cli spike 跑一次最小验证。

3. 调用

# 健康检查
curl -s http://127.0.0.1:8765/health

# 翻译(自动检测语言,英文→中文)
curl -s -X POST http://127.0.0.1:8765/translate -H 'Content-Type: application/json' \
  -d '{"text": "The quick brown fox jumps over the lazy dog."}'

# 流式(SSE)
curl -s -N -X POST http://127.0.0.1:8765/translate -H 'Content-Type: application/json' \
  -d '{"text": "今天天气真好", "stream": true}'

# /v1/chat/completions 接口
curl -s -X POST http://127.0.0.1:8765/v1/chat/completions -H 'Content-Type: application/json' \
  -d '{"messages": [{"role": "user", "content": "Hello world"}]}'

PopClip 集成

  1. 保持 gemma-trans-cli serve(或 app)运行;用 app 时在菜单栏确认"本地 API"开启(默认已开)
  2. 在 Finder 中双击 popclip/GemmaTrans.popclipext 目录,PopClip 会提示安装
  3. 任意 app 选中文字 → 点击 PopClip 弹条中的 GemmaTrans 图标 → 顶部显示译文

PopClip 也可以用任何接受 /v1/chat/completions 的扩展指向 http://127.0.0.1:8765/v1(API key 随意填)。

API 文档

POST /translate

字段 类型 说明
text string 必填,待翻译文本(超长截断;上限随机器自动调优,16GB 机器默认 1500 字符,可在设置中手动覆盖)
target string? 目标语言 BCP-47 码(如 enzh-Hans);缺省走智能双向
stream bool? true 时返回 SSE 流

响应:{"translation": "...", "detected": "en", "target": "zh-Hans", "truncated": false}

SSE 流格式:若干 data: {"delta": "..."} → 一条 data: {"translation": 全文, ...}data: [DONE]

POST /v1/chat/completions

兼容 /v1/chat/completions 请求格式(含 stream: true SSE),方便接受该格式的工具直连本地引擎。取最后一条 user 消息按智能双向翻译;model 字段与 system 消息被忽略——这是翻译器,不是通用聊天,且全程本地,不连任何外部服务。

GET /health

{"status": "ready"}(200)或 {"status": "loading"}(503)。

错误码:400(空文本/无效 JSON)、503(模型未加载 / 引擎忙超时 30s)、500(引擎错误,error 字段含详情,常见于系统内存压力过高时 GPU 分配失败)。

引擎与服务日志:~/Library/Logs/GemmaTrans/gemmatrans.log(GUI app 无 stderr,排障看这里)。

API 可在菜单栏/设置中即时开关,无需重启。关闭后划词翻译不受影响——划词是进程内调用,不走端口。端口被其他程序占用时菜单会显示"API 失败",划词照常可用;若占用者是另一个 GemmaTrans 实例(如 CLI serve),app 会拒绝启动以避免加载两份模型。

Menu bar app(划词翻译)

brew install xcodegen   # 首次
cd App && xcodegen generate
xcodebuild -project GemmaTrans.xcodeproj -scheme GemmaTrans -configuration Debug -derivedDataPath build build
open build/Build/Products/Debug/GemmaTrans.app
  • 启动后弹出主窗口(含模型下载进度),同时 menu bar 出现 💬 图标;模型加载完成(约 15s)后图标变实心,并在 127.0.0.1:8765 提供 API(app 与 CLI serve 二者跑一个即可,同时跑会端口冲突)
  • 三种译法(均无需任何系统权限):
    • 主窗口里粘贴/输入文字 → 点「翻译」流式显示译文
    • 任意 app 选中文字 → 按 ⌥⌘T 一键翻译选中内容,无需先复制;也可从「服务」菜单点「Translate with GemmaTrans」
      • 这是 macOS「服务」快捷键,默认 ⌥⌘T。macOS 不一定会自动启用声明的默认值——若按了没反应,去 系统设置 › 键盘 › 键盘快捷键 › 服务 › 文本,勾选并确认 Translate with GemmaTrans 的快捷键(一次性,之后长期生效;也可在这里改键)
    • 复制文字后按 ⌥D → 翻译剪贴板内容(鼠标旁浮窗流式显示,Esc 关闭,可一键复制)
  • 菜单"设置…"可改目标语言、API 端口和剪贴板热键
  • 设置"性能"区默认按机器内存自动配置 KV cache 与输入上限(加载时还会按当前可用内存降档),也可手动覆盖

架构

GemmaTransKit     核心库:MLX-Swift 引擎封装、语言检测(NaturalLanguage)、提示词、按内存自动调优
GemmaTransServer  HTTP 层:FlyingFox,/translate + /v1/chat/completions + SSE
gemma-trans-cli   命令行:spike / serve