TYH-labs/unsloth-buddy

{   "name": "unsloth-buddy",   "owner": { "name": "Gaslamp", "url": "https://gaslamp.dev" },

"name": "unsloth-buddy",   "owner": { "name": "Gaslamp", "url": "https://gaslamp.dev" },   "metadata": {

"owner": { "name": "Gaslamp", "url": "https://gaslamp.dev" },

"description": "Agentic LLM fine-tuning skill for Claude Code — Unsloth on NVIDIA, mlx-tune on Apple Silicon.",     "version": "1.0.5",     "repository": "https://github.com/TYH-labs/unsloth-buddy",

"version": "1.0.5",     "repository": "https://github.com/TYH-labs/unsloth-buddy",     "documentation": "https://gaslamp.dev/unsloth"

"repository": "https://github.com/TYH-labs/unsloth-buddy",

"documentation": "https://gaslamp.dev/unsloth"

> This file is read by [OpenClaw](https://github.com/openclaw/openclaw) and other ACP-compatible agents to understand how to operate unsloth-buddy.

<p align="center">   <a href="https://github.com/TYH-labs/unsloth-buddy"><img src="https://img.shields.io/github/stars/TYH-labs/unsloth-buddy?style=flat&logo=github&color=181717&logoColor=white" alt="GitHub" /></a>

<a href="https://github.com/TYH-labs/unsloth-buddy"><img src="https://img.shields.io/github/stars/TYH-labs/unsloth-buddy?style=flat&logo=github&color=181717&logoColor=white" alt="GitHub" /></a>

<a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="MIT License" /></a>

<a href="#quick-start"><img src="https://img.shields.io/badge/Python-3.10%2B-3776AB?logo=python&logoColor=white" alt="Python 3.10+" /></a>

<a href="https://gaslamp.dev/unsloth"><img src="https://img.shields.io/badge/%F0%9F%94%A5%20Gaslamp-Compatible-ff6b00?logoColor=white" alt="Gaslamp Compatible" /></a>

<a href="#openclaw"><img src="https://img.shields.io/badge/%F0%9F%A6%9E%20OpenClaw-Compatible-ff4444" alt="OpenClaw Compatible" /></a>

<a href="#quick-start"><img src="https://img.shields.io/badge/%F0%9F%A4%96%20Agent-Claude%20Code%20%2F%20Codex%20%2F%20Gemini-8b5cf6" alt="Agent Compatible" /></a>

<a href="https://discord.gg/mZe4mbCQ6a"><img src="https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&logoColor=white" alt="Discord" /></a>

<a href="#local-deploy"><img src="https://img.shields.io/badge/Backend-Unsloth%20%7C%20MLX%20%7C%20llama.cpp-00b4d8" alt="Backend: Unsloth / MLX / llama.cpp" /></a>

<p align="center">   <a href="#quick-start"><img src="https://img.shields.io/badge/Try%20It-1%20minute-black?style=for-the-badge" alt="Try It" /></a>

<a href="#quick-start"><img src="https://img.shields.io/badge/Try%20It-1%20minute-black?style=for-the-badge" alt="Try It" /></a>

<a href="demos/"><img src="https://img.shields.io/badge/Demos-Examples-6e40c9?style=for-the-badge" alt="Demos" /></a>

<a href="SKILL.md"><img src="https://img.shields.io/badge/10%2B%20Features-Details-0969da?style=for-the-badge" alt="Features" /></a>

<p align="center">   <a href="https://youtu.be/wG28uxDGjHE"><img src="https://img.shields.io/badge/▶%20Demo-YouTube-FF0000?style=for-the-badge&logo=youtube&logoColor=white" alt="YouTube Demo" /></a>

<a href="https://youtu.be/wG28uxDGjHE"><img src="https://img.shields.io/badge/▶%20Demo-YouTube-FF0000?style=for-the-badge&logo=youtube&logoColor=white" alt="YouTube Demo" /></a>

<a href="https://www.bilibili.com/video/BV1VWAFzmECy/"><img src="https://img.shields.io/badge/▶%20演示-Bilibili-00A1D6?style=for-the-badge&logo=bilibili&logoColor=white" alt="Bilibili Demo" /></a>

Runs on NVIDIA GPUs via [Unsloth](https://github.com/unslothai/unsloth), natively on Apple Silicon via [mlx-tune](https://github.com/ml-explore/mlx-lm), and on free cloud GPUs via [colab-mcp](https://github.com/googlecolab/colab-mcp). Part 

gemini extensions install https://github.com/TYH-labs/unsloth-buddy --consent

**Any agent supporting the [Agent Skills](https://agentskills.io/) standard**

git clone https://github.com/TYH-labs/unsloth-buddy.git .agents/skills/unsloth-buddy

Every local training run automatically opens a real-time dashboard at **http://localhost:8080/**:

The demo builder is part of the [Gaslamp](https://gaslamp.dev/) platform's presentation toolkit. We've simplified it for unsloth-buddy with two built-in themes and automatic domain-specific color customization:

This runs the full pipeline: quantize → benchmark → start an OpenAI-compatible server → open the Gaslamp Chat WebUI at `http://localhost:8081/`. Individual subcommands are also available:

Requires [llama.cpp](https://github.com/ggml-org/llama.cpp) — installed automatically via `llamacpp.py install`.

uv python install 3.13    claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

`unsloth-buddy` works standalone or as part of a larger [Gaslamp](https://gaslamp.dev/) project — an agentic platform that orchestrates the full ML lifecycle from research to training to deployment. When called via Gaslamp, the project dire

[gaslamp.dev/unsloth](https://gaslamp.dev/unsloth) — [gaslamp.dev](https://gaslamp.dev/)

**unsloth-buddy is an [OpenClaw](https://github.com/openclaw/openclaw)-compatible skill.** Share the repo URL with OpenClaw, describe what you want to fine-tune — it reads `AGENTS.md`, understands the workflow, and runs everything automatic

1. Share https://github.com/TYH-labs/unsloth-buddy with OpenClaw

- **2026-04-14** — **Self-evolving memory** (Phase 7 + global inject): after each project, the agent synthesizes lessons, model-specific gotchas, and reusable scenario recipes into `~/.gaslamp/`. Every new project injects a frozen snapshot 

- **2026-04-12** — Added **llama.cpp local deploy** (Phase 6.5): after GGUF export, if llama.cpp is installed, the agent offers a one-command pipeline — quantize → benchmark → serve + open the Gaslamp Chat WebUI (`templates/chat_ui.html`). 

- **2026-04-10** — Added native **Vision SFT for Apple Silicon**: Integrated [mlx-vlm](https://github.com/Blaizzy/mlx-vlm) to support multimodal fine-tuning (e.g. Gemma 4 Vision, Qwen2.5-VL) on M-series chips. Added `scripts/unsloth_mlx_vis

- **2026-04-09** — Demo Builder improvements: auto-resolves conceptual/movie keywords (e.g. "matrix" → nvidia, "star wars" → spacex) to the best-fit brand before calling the design search script; distinguishes shallow vs. deep DESIGN.md ove

- **2026-04-04** — Added Demo Builder (Phase 5.5): after evaluation, generates a static HTML demo page showing base vs fine-tuned outputs side-by-side. Two themes (crisp-light, dark-signal) with automatic domain-specific accent colors. No s

- **2026-03-21** — Enhanced training dashboard: task-aware panels (SFT/DPO/GRPO/Vision), GPU memory breakdown (baseline vs LoRA vs total), GRPO reward ± std and KL divergence charts, DPO chosen/rejected reward and KL charts, epoch tracking,

- **2026-03-18** — Added Google Colab training support via [colab-mcp](https://github.com/googlecolab/colab-mcp): free T4/L4/A100 GPU access from Claude Code, background-thread training with live polling, and adapter download workflow.

<p align="center">   <a href="https://github.com/TYH-labs/unsloth-buddy"><img src="https://img.shields.io/github/stars/TYH-labs/unsloth-buddy?style=flat&logo=github&color=181717&logoColor=white" alt="GitHub" /></a>

<a href="https://github.com/TYH-labs/unsloth-buddy"><img src="https://img.shields.io/github/stars/TYH-labs/unsloth-buddy?style=flat&logo=github&color=181717&logoColor=white" alt="GitHub" /></a>

<a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="MIT License" /></a>

<a href="#快速开始"><img src="https://img.shields.io/badge/Python-3.10%2B-3776AB?logo=python&logoColor=white" alt="Python 3.10+" /></a>

<a href="https://gaslamp.dev/unsloth"><img src="https://img.shields.io/badge/%F0%9F%94%A5%20Gaslamp-Compatible-ff6b00?logoColor=white" alt="Gaslamp Compatible" /></a>

<a href="#openclaw"><img src="https://img.shields.io/badge/%F0%9F%A6%9E%20OpenClaw-Compatible-ff4444" alt="OpenClaw Compatible" /></a>

<a href="#快速开始"><img src="https://img.shields.io/badge/%F0%9F%A4%96%20Agent-Claude%20Code%20%2F%20Codex%20%2F%20Gemini-8b5cf6" alt="Agent Compatible" /></a>

<a href="https://discord.gg/mZe4mbCQ6a"><img src="https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&logoColor=white" alt="Discord" /></a>

<a href="#本地部署"><img src="https://img.shields.io/badge/后端-Unsloth%20%7C%20MLX%20%7C%20llama.cpp-00b4d8" alt="后端：Unsloth / MLX / llama.cpp" /></a>

<p align="center">   <a href="#快速开始"><img src="https://img.shields.io/badge/立即体验-1 分钟-black?style=for-the-badge" alt="立即体验" /></a>

<a href="#快速开始"><img src="https://img.shields.io/badge/立即体验-1 分钟-black?style=for-the-badge" alt="立即体验" /></a>

<a href="demos/"><img src="https://img.shields.io/badge/演示案例-示例-6e40c9?style=for-the-badge" alt="演示案例" /></a>

<a href="SKILL.md"><img src="https://img.shields.io/badge/10%2B%20功能-功能详情-0969da?style=for-the-badge" alt="10+ 功能" /></a>

<p align="center">   <a href="https://www.bilibili.com/video/BV1VWAFzmECy/"><img src="https://img.shields.io/badge/▶%20演示-Bilibili-00A1D6?style=for-the-badge&logo=bilibili&logoColor=white" alt="Bilibili 演示" /></a>

<a href="https://www.bilibili.com/video/BV1VWAFzmECy/"><img src="https://img.shields.io/badge/▶%20演示-Bilibili-00A1D6?style=for-the-badge&logo=bilibili&logoColor=white" alt="Bilibili 演示" /></a>

<a href="https://youtu.be/wG28uxDGjHE"><img src="https://img.shields.io/badge/▶%20Demo-YouTube-FF0000?style=for-the-badge&logo=youtube&logoColor=white" alt="YouTube Demo" /></a>

支持通过 [Unsloth](https://github.com/unslothai/unsloth) 在 NVIDIA GPU 上运行，也支持通过 [mlx-tune](https://github.com/ml-explore/mlx-lm) 在 Apple Silicon 上原生运行。隶属于 [Gaslamp](https://gaslamp.dev/) AI 开发平台 — [文档](https://gaslamp.dev/unsloth)。

gemini extensions install https://github.com/TYH-labs/unsloth-buddy --consent

**任何支持 [Agent Skills](https://agentskills.io/) 标准的 Agent**

git clone https://github.com/TYH-labs/unsloth-buddy.git .agents/skills/unsloth-buddy

每次本地训练都会自动在 **http://localhost:8080/** 打开实时看板：

演示生成器是 [Gaslamp](https://gaslamp.dev/) 平台展示工具的一部分。我们为 unsloth-buddy 做了简化，内置两套主题并支持基于领域的自动配色：

将运行完整流水线：量化 → 基准测试 → 启动 OpenAI 兼容服务器 → 在浏览器中打开 Gaslamp Chat WebUI（`http://localhost:8081/`）。也可单独使用各子命令：

需要安装 [llama.cpp](https://github.com/ggml-org/llama.cpp) — 可通过 `llamacpp.py install` 自动安装。

uv python install 3.13    claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

`unsloth-buddy` 可独立使用，也可作为 [Gaslamp](https://gaslamp.dev/) 大型项目的一部分运行 — Gaslamp 是一个协调从研究到训练再到部署整个 ML 生命周期的 Agentic 平台。通过 Gaslamp 调用时，项目目录和状态在各个技能之间共享，结果会自动传递到下一阶段。

[gaslamp.dev/unsloth](https://gaslamp.dev/unsloth) — [gaslamp.dev](https://gaslamp.dev/)

**unsloth-buddy 是一个 [OpenClaw](https://github.com/openclaw/openclaw) 兼容技能。** 将仓库 URL 分享给 OpenClaw，描述你想微调的内容 — 它会读取 `AGENTS.md`，理解流程，并自动运行一切。

1. 将 https://github.com/TYH-labs/unsloth-buddy 分享给 OpenClaw

- **2026-04-14** — **自进化记忆**（第 7 阶段 + 全局注入）：每次项目完成后，Agent 将经验、模型特有注意事项与可复用场景配方提炼至 `~/.gaslamp/`。每个新项目启动时自动注入冻结快照，静默应用历史知识。实现了代理记忆研究中的冻结快照模式。详见 `ref/self_evolve_plan.md`。 - **2026-04-12** — 新增 **llama.cpp 本地部署**（第 6.5 阶段）：GGUF 导出完成后，若检测到 lla

- **2026-04-12** — 新增 **llama.cpp 本地部署**（第 6.5 阶段）：GGUF 导出完成后，若检测到 llama.cpp，Agent 将提供一键流水线 — 量化 → 基准测试 → 启动服务 + 打开 Gaslamp Chat WebUI（`templates/chat_ui.html`）。`scripts/llamacpp.py` 提供 7 个子命令（`install`、`quantize`、`bench`、`ppl`、`serve`、`cha

- **2026-04-10** — 新增 Apple Silicon 原生**视觉 SFT 支持**：集成 [mlx-vlm](https://github.com/Blaizzy/mlx-vlm) 以支持在 M 系列芯片上进行多模态微调（如 Gemma 4 Vision、Qwen2.5-VL）。新增 `scripts/unsloth_mlx_vision_example.py` 训练模板和 `mlx_eval_vision_template.py` 视觉对比评估脚本。演示

- **2026-04-09** — 演示生成器改进：自动将概念/电影类关键词（如 "matrix" → nvidia，"star wars" → spacex）解析并映射至最匹配的品牌；区分浅层与深层 DESIGN.md 覆盖 —— 深层覆盖（如全黑或局部大面积结构性分割布局）将跳过 CSS 注入点，从头开始写入演示文件。将 `scripts/search_design.py` 补充至技能脚本汇总，即使环境内无 `npx` 也可直接获取品牌设计模板。 - **2026-04

- **2026-04-04** — 新增演示生成器（第 5.5 阶段）：评估完成后生成静态 HTML 演示页面，并排展示基础模型与微调模型输出。内置两套主题（crisp-light、dark-signal），支持基于领域的自动强调色。无需服务器 — 在任意浏览器中直接打开。属于 [Gaslamp](https://gaslamp.dev/) 展示工具的简化版本。访谈从 5 点合同简化为 2 个问题（任务 + 数据），同时捕获用户领域/受众以用于演示主题选择。

- **2026-03-21** — 增强训练看板：任务感知面板（SFT/DPO/GRPO/Vision）、GPU 内存分解（基线 vs LoRA vs 总量）、GRPO 奖励 ± 标准差及 KL 散度图表、DPO 选中/拒绝奖励及 KL 图表、轮次追踪、训练完成摘要横幅、终端 DPO/GRPO 2×2 布局，以及新增 `scripts/demo_server.py` 无需 GPU 即可预览所有面板的模拟服务器。 - **2026-03-19** — 新增终端训练看板（`sc

- **2026-03-18** — 新增通过 [colab-mcp](https://github.com/googlecolab/colab-mcp) 的 Google Colab 云端训练支持：可在 Claude Code 中直接使用免费 T4/L4/A100 GPU，支持后台线程训练、实时轮询进度及适配器下载流程。

<p align="center">   <a href="https://github.com/TYH-labs/unsloth-buddy"><img src="https://img.shields.io/github/stars/TYH-labs/unsloth-buddy?style=flat&logo=github&color=181717&logoColor=white" alt="GitHub" /></a>

<a href="https://github.com/TYH-labs/unsloth-buddy"><img src="https://img.shields.io/github/stars/TYH-labs/unsloth-buddy?style=flat&logo=github&color=181717&logoColor=white" alt="GitHub" /></a>

<a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="MIT License" /></a>

<a href="#快速開始"><img src="https://img.shields.io/badge/Python-3.10%2B-3776AB?logo=python&logoColor=white" alt="Python 3.10+" /></a>

<a href="https://gaslamp.dev/unsloth"><img src="https://img.shields.io/badge/%F0%9F%94%A5%20Gaslamp-Compatible-ff6b00?logoColor=white" alt="Gaslamp Compatible" /></a>

<a href="#openclaw"><img src="https://img.shields.io/badge/%F0%9F%A6%9E%20OpenClaw-Compatible-ff4444" alt="OpenClaw Compatible" /></a>

<a href="#快速開始"><img src="https://img.shields.io/badge/%F0%9F%A4%96%20Agent-Claude%20Code%20%2F%20Codex%20%2F%20Gemini-8b5cf6" alt="Agent Compatible" /></a>

<a href="https://discord.gg/mZe4mbCQ6a"><img src="https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&logoColor=white" alt="Discord" /></a>

<a href="#本地部署"><img src="https://img.shields.io/badge/後端-Unsloth%20%7C%20MLX%20%7C%20llama.cpp-00b4d8" alt="後端：Unsloth / MLX / llama.cpp" /></a>

<p align="center">   <a href="#快速開始"><img src="https://img.shields.io/badge/立即體驗-1 分鐘-black?style=for-the-badge" alt="立即體驗" /></a>

<a href="#快速開始"><img src="https://img.shields.io/badge/立即體驗-1 分鐘-black?style=for-the-badge" alt="立即體驗" /></a>

<a href="demos/"><img src="https://img.shields.io/badge/示範案例-範例-6e40c9?style=for-the-badge" alt="示範案例" /></a>

<a href="SKILL.md"><img src="https://img.shields.io/badge/10%2B%20功能-功能詳情-0969da?style=for-the-badge" alt="10+ 功能" /></a>

<p align="center">   <a href="https://www.bilibili.com/video/BV1VWAFzmECy/"><img src="https://img.shields.io/badge/▶%20演示-Bilibili-00A1D6?style=for-the-badge&logo=bilibili&logoColor=white" alt="Bilibili 演示" /></a>

<a href="https://www.bilibili.com/video/BV1VWAFzmECy/"><img src="https://img.shields.io/badge/▶%20演示-Bilibili-00A1D6?style=for-the-badge&logo=bilibili&logoColor=white" alt="Bilibili 演示" /></a>

<a href="https://youtu.be/wG28uxDGjHE"><img src="https://img.shields.io/badge/▶%20Demo-YouTube-FF0000?style=for-the-badge&logo=youtube&logoColor=white" alt="YouTube Demo" /></a>

支援透過 [Unsloth](https://github.com/unslothai/unsloth) 在 NVIDIA GPU 上執行，也支援透過 [mlx-tune](https://github.com/ml-explore/mlx-lm) 在 Apple Silicon 上原生執行。隸屬於 [Gaslamp](https://gaslamp.dev/) AI 開發平台 — [文件](https://gaslamp.dev/unsloth)。

gemini extensions install https://github.com/TYH-labs/unsloth-buddy --consent

**任何支援 [Agent Skills](https://agentskills.io/) 標準的 Agent**

git clone https://github.com/TYH-labs/unsloth-buddy.git .agents/skills/unsloth-buddy

每次本地訓練都會自動在 **http://localhost:8080/** 開啟即時儀表板：

展示生成器是 [Gaslamp](https://gaslamp.dev/) 平台展示工具的一部分。我們為 unsloth-buddy 做了簡化，內建兩套主題並支援基於領域的自動配色：

將執行完整流水線：量化 → 基準測試 → 啟動 OpenAI 相容伺服器 → 在瀏覽器中開啟 Gaslamp Chat WebUI（`http://localhost:8081/`）。也可單獨使用各子指令：

需要安裝 [llama.cpp](https://github.com/ggml-org/llama.cpp) — 可透過 `llamacpp.py install` 自動安裝。

uv python install 3.13    claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

`unsloth-buddy` 可獨立使用，也可作為 [Gaslamp](https://gaslamp.dev/) 大型專案的一部分執行 — Gaslamp 是一個協調從研究到訓練再到部署整個 ML 生命週期的 Agentic 平台。透過 Gaslamp 呼叫時，專案目錄和狀態在各個技能之間共享，結果會自動傳遞到下一階段。

[gaslamp.dev/unsloth](https://gaslamp.dev/unsloth) — [gaslamp.dev](https://gaslamp.dev/)

**unsloth-buddy 是一個 [OpenClaw](https://github.com/openclaw/openclaw) 相容技能。** 將儲存庫 URL 分享給 OpenClaw，描述你想微調的內容 — 它會讀取 `AGENTS.md`，理解流程，並自動執行一切。

1. 將 https://github.com/TYH-labs/unsloth-buddy 分享給 OpenClaw

- **2026-04-14** — **自進化記憶**（第 7 階段 + 全域注入）：每次專案完成後，Agent 將經驗、模型特有注意事項與可重複使用場景配方提煉至 `~/.gaslamp/`。每個新專案啟動時自動注入凍結快照，靜默套用歷史知識。實現了代理記憶研究中的凍結快照模式。詳見 `ref/self_evolve_plan.md`。 - **2026-04-12** — 新增 **llama.cpp 本地部署**（第 6.5 階段）：GGUF 匯出完成後，若偵測到 l

- **2026-04-12** — 新增 **llama.cpp 本地部署**（第 6.5 階段）：GGUF 匯出完成後，若偵測到 llama.cpp，Agent 將提供一鍵流水線 — 量化 → 基準測試 → 啟動服務 + 開啟 Gaslamp Chat WebUI（`templates/chat_ui.html`）。`scripts/llamacpp.py` 提供 7 個子指令（`install`、`quantize`、`bench`、`ppl`、`serve`、`cha

- **2026-04-10** — 新增 Apple Silicon 原生**視覺 SFT 支援**：整合 [mlx-vlm](https://github.com/Blaizzy/mlx-vlm) 以支援在 M 系列晶片上進行多模態微調（如 Gemma 4 Vision、Qwen2.5-VL）。新增 `scripts/unsloth_mlx_vision_example.py` 訓練範本和 `mlx_eval_vision_template.py` 視覺對比評估腳本。展示

- **2026-04-09** — 展示生成器改進：在呼叫設計搜尋腳本前，自動將概念/電影類關鍵字（如 "matrix" → nvidia，"star wars" → spacex）解析並對應至最匹配的品牌；區分淺層與深層 DESIGN.md 覆蓋 —— 深層覆蓋（如全黑或局部大面積結構性分割佈局）將跳過 CSS 注入點，從頭開始寫入展示檔案。將 `scripts/search_design.py` 補充至技能資源，即使環境內無 `npx` 也可直接獲取品牌設計範本。 - 

- **2026-04-04** — 新增展示生成器（第 5.5 階段）：評估完成後產生靜態 HTML 展示頁面，並排展示基礎模型與微調模型輸出。內建兩套主題（crisp-light、dark-signal），支援基於領域的自動強調色。無需伺服器 — 在任意瀏覽器中直接開啟。屬於 [Gaslamp](https://gaslamp.dev/) 展示工具的簡化版本。訪談從 5 點合約簡化為 2 個問題（任務 + 資料），同時擷取使用者領域/受眾以用於展示主題選擇。

- **2026-03-21** — 增強訓練儀表板：任務感知面板（SFT/DPO/GRPO/Vision）、GPU 記憶體分解（基線 vs LoRA vs 總量）、GRPO 獎勵 ± 標準差及 KL 散度圖表、DPO 選取/拒絕獎勵及 KL 圖表、輪次追蹤、訓練完成摘要橫幅、終端 DPO/GRPO 2×2 配置，以及新增 `scripts/demo_server.py` 無需 GPU 即可預覽所有面板的模擬伺服器。 - **2026-03-19** — 新增終端訓練儀表板（

- **2026-03-18** — 新增透過 [colab-mcp](https://github.com/googlecolab/colab-mcp) 的 Google Colab 雲端訓練支援：可在 Claude Code 中直接使用免費 T4/L4/A100 GPU，支援背景執行緒訓練、即時輪詢進度及適配器下載流程。

version: "1.0.3"   category: machine-learning   repository: https://github.com/TYH-labs/unsloth-buddy

category: machine-learning   repository: https://github.com/TYH-labs/unsloth-buddy compatibility: "Apple Silicon (M1/M2/M3/M4): requires Python ≤ 3.12 and mlx-tune (not CUDA). Linux/WSL with NVIDIA GPU: CUDA 11.8, 12.1, or 12.4+, Python 3.1

repository: https://github.com/TYH-labs/unsloth-buddy

# 2. Add colab-mcp to Claude Code claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

"colab-mcp": {   "command": "uvx",   "args": ["--from", "git+https://github.com/googlecolab/colab-mcp",

"command": "uvx",   "args": ["--from", "git+https://github.com/googlecolab/colab-mcp",            "--python", "3.13", "colab-mcp"],

"args": ["--from", "git+https://github.com/googlecolab/colab-mcp",

1. Tell the user to open a new notebook at https://colab.research.google.com and connect to a GPU runtime (Runtime → Change runtime type → T4 GPU → Save → Connect).

trainer.train()     ```     Dashboard serves at **http://localhost:8080/** with loss, learning rate, val loss, Peak mem (GB), tokens/sec.

```     Dashboard serves at **http://localhost:8080/** with loss, learning rate, val loss, Peak mem (GB), tokens/sec. - **Terminal Dashboard** — ALWAYS install and present this to the user before starting training, regardless of response la

Dashboard serves at **http://localhost:8080/** with loss, learning rate, val loss, Peak mem (GB), tokens/sec.

- Remind the user of live monitoring options (the HTTP server starts automatically with training):     - **Web dashboard**: Open **http://localhost:8080/** in a browser for the live interactive dashboard.

- **Web dashboard**: Open **http://localhost:8080/** in a browser for the live interactive dashboard.

wget -qO- https://raw.githubusercontent.com/unslothai/unsloth/main/unsloth/_auto_install.py | python -

wget -qO- https://raw.githubusercontent.com/unslothai/unsloth/main/unsloth/_auto_install.py | python -

Guide the user to: 1. Create environment: `conda create --name unsloth_env python==3.12 -y` & `conda activate unsloth_env` 2. Install PyTorch for their CUDA version (e.g. `pip3 install torch torchvision torchaudio --index-url https://downlo

1. Create environment: `conda create --name unsloth_env python==3.12 -y` & `conda activate unsloth_env` 2. Install PyTorch for their CUDA version (e.g. `pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/

2. Install PyTorch for their CUDA version (e.g. `pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121`)

Tell them to access Jupyter Lab at `http://localhost:8888`.

"colab-mcp": {       "command": "uvx",       "args": ["git+https://github.com/googlecolab/colab-mcp"],

"command": "uvx",       "args": ["git+https://github.com/googlecolab/colab-mcp"],       "timeout": 30000

"args": ["git+https://github.com/googlecolab/colab-mcp"],

**Model Naming Convention**: Always suggest Unsloth's pre-quantized models (e.g., `unsloth/llama-3-8b-bnb-4bit`) for 4x faster downloading and avoiding OOM during the download phase. Browse the full catalog at https://unsloth.ai/docs/get-st

{"messages": [     {"role": "user", "content": [         {"type": "image", "image": "https://example.com/image.jpg"},

{"role": "user", "content": [         {"type": "image", "image": "https://example.com/image.jpg"},         {"type": "text",  "text": "Describe this image."}

{"type": "image", "image": "https://example.com/image.jpg"},

### 7. Gradient accumulation bug - **Fix**: Use Unsloth's patched trainers. Standard HuggingFace trainers have a known gradient accumulation bug that Unsloth fixes automatically. See: https://unsloth.ai/blog/gradient

- **Fix**: Use Unsloth's patched trainers. Standard HuggingFace trainers have a known gradient accumulation bug that Unsloth fixes automatically. See: https://unsloth.ai/blog/gradient

- [Unsloth Documentation](https://unsloth.ai/docs)

- [Model Catalog](https://unsloth.ai/docs/get-started/unsloth-model-catalog)

- [Fine-tuning Guide](https://unsloth.ai/docs/get-started/fine-tuning-llms-guide)

- [RL Guide](https://unsloth.ai/docs/get-started/reinforcement-learning-rl-guide)

- [Vision Guide](https://unsloth.ai/docs/basics/vision-fine-tuning)

- [TTS Guide](https://unsloth.ai/docs/basics/text-to-speech-tts-fine-tuning)

- [Saving to GGUF](https://unsloth.ai/docs/basics/inference-and-deployment/saving-to-gguf)

- [vLLM Deployment](https://unsloth.ai/docs/basics/inference-and-deployment/vllm-guide)

- [All Notebooks](https://unsloth.ai/docs/get-started/unsloth-notebooks)

Every example is a single file. Hand it to any agent running [unsloth-buddy](https://gaslamp.dev/unsloth) and the entire project reproduces end-to-end: data download, environment setup, training, and evaluation.

/install-plugin https://github.com/TYH-labs/unsloth-buddy

每个示例只有一个文件。将它交给运行 [unsloth-buddy](https://gaslamp.dev/unsloth) 的 Agent，整个项目即可端到端复现：数据下载、环境配置、训练和评估。

/install-plugin https://github.com/TYH-labs/unsloth-buddy

每個範例只有一個檔案。將它交給執行 [unsloth-buddy](https://gaslamp.dev/unsloth) 的 Agent，整個專案即可端到端重現：資料下載、環境設定、訓練與評估。

/install-plugin https://github.com/TYH-labs/unsloth-buddy

# gaslamp.md — Roadbook for `gemma4_htr_sft` # Powered by Gaslamp · https://gaslamp.dev

# Powered by Gaslamp · https://gaslamp.dev

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

@import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

# gaslamp.md — Roadbook for `qwen_chip2_sft` # Powered by Gaslamp · https://gaslamp.dev

# Powered by Gaslamp · https://gaslamp.dev

**Source:** `laion/OIG` — `unified_chip2.jsonl` via HuggingFace **URL:** `https://huggingface.co/datasets/laion/OIG/resolve/main/unified_chip2.jsonl`

**URL:** `https://huggingface.co/datasets/laion/OIG/resolve/main/unified_chip2.jsonl`

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

@import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

@import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

@import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Executable file

Executable file

Executable file

Executable file

Executable file

Executable file

[Credential-related source omitted]

Executable file

#!/usr/bin/env python3

[Credential-related source omitted]

Executable file

Executable file

Executable file

Executable file

Executable file

Executable file

2. Add colab-mcp to Claude Code:      claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

claude mcp add colab-mcp -- uvx --from git+https://github.com/googlecolab/colab-mcp --python 3.13 colab-mcp

"colab-mcp": {        "command": "uvx",        "args": ["--from", "git+https://github.com/googlecolab/colab-mcp",

"command": "uvx",        "args": ["--from", "git+https://github.com/googlecolab/colab-mcp",                 "--python", "3.13", "colab-mcp"],

"args": ["--from", "git+https://github.com/googlecolab/colab-mcp",

4. Open a new Colab notebook at https://colab.research.google.com

Executable file

_PAYLOAD = make_payload(args.task, args.hardware)     httpd = HTTPServer(("localhost", args.port), Handler)     print(f"🚀  Demo dashboard at  http://localhost:{args.port}/  (task={args.task}, hardware={args.hardware})")

httpd = HTTPServer(("localhost", args.port), Handler)     print(f"🚀  Demo dashboard at  http://localhost:{args.port}/  (task={args.task}, hardware={args.hardware})")     print("    Press Ctrl+C to stop.")

print(f"🚀  Demo dashboard at  http://localhost:{args.port}/  (task={args.task}, hardware={args.hardware})")

Executable file

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

[Credential-related source omitted]

Executable file

Executable file

time.sleep(0.5)             print("\n" + "=" * 60)             print(f"🚀 Gaslamp Live Training Dashboard: http://localhost:{self.port}/")

print("\n" + "=" * 60)             print(f"🚀 Gaslamp Live Training Dashboard: http://localhost:{self.port}/")             print("=" * 60 + "\n")

print(f"🚀 Gaslamp Live Training Dashboard: http://localhost:{self.port}/")

Executable file

#!/usr/bin/env python3

os.makedirs(build_dir, exist_ok=True)             _run(["git", "clone", "--depth=1",                   "https://github.com/ggml-org/llama.cpp.git", build_dir],

_run(["git", "clone", "--depth=1",                   "https://github.com/ggml-org/llama.cpp.git", build_dir],                  check=False)

"https://github.com/ggml-org/llama.cpp.git", build_dir],

else:         print(f"⚠<U+FE0F>  Auto-install not supported on {system}.")         print("    Download from: https://github.com/ggml-org/llama.cpp/releases")

print(f"⚠<U+FE0F>  Auto-install not supported on {system}.")         print("    Download from: https://github.com/ggml-org/llama.cpp/releases")         sys.exit(1)

print("    Download from: https://github.com/ggml-org/llama.cpp/releases")

else:         print("❌  Installation failed. Try manual install:")         print("    https://github.com/ggml-org/llama.cpp/releases")

print("❌  Installation failed. Try manual install:")         print("    https://github.com/ggml-org/llama.cpp/releases")         sys.exit(1)

print("    https://github.com/ggml-org/llama.cpp/releases")

try:             import urllib.request             resp = urllib.request.urlopen(f"http://localhost:{args.port}/health", timeout=2)

import urllib.request             resp = urllib.request.urlopen(f"http://localhost:{args.port}/health", timeout=2)             if resp.status == 200:

resp = urllib.request.urlopen(f"http://localhost:{args.port}/health", timeout=2)

print(f"✅  Server ready!")     print(f"    API:    http://localhost:{args.port}/v1/chat/completions")

print(f"    API:    http://localhost:{args.port}/v1/chat/completions")

print(f"    WebUI:  http://localhost:{args.port}/")

webbrowser.open(f"http://localhost:{args.port}/")

Executable file

Executable file

Executable file

Dashboard → http://localhost:8080/

"""     Context manager that wraps mlx-tune's trainer.train() to provide a live     Gaslamp dashboard at http://localhost:{port}/.

Gaslamp dashboard at http://localhost:{port}/.

time.sleep(0.3)             print(f"\n{'='*60}")             print(f"🚀 Gaslamp Live Training Dashboard: http://localhost:{self.port}/")

print(f"\n{'='*60}")             print(f"🚀 Gaslamp Live Training Dashboard: http://localhost:{self.port}/")             print(f"{'='*60}\n")

print(f"🚀 Gaslamp Live Training Dashboard: http://localhost:{self.port}/")

Executable file

[Credential-related source omitted]

print("[train] Starting GRPO training...")     print("[train] Dashboard → http://localhost:8080/")

print("[train] Dashboard → http://localhost:8080/")

Executable file

Executable file

def fetch_json(url):     req = urllib.request.Request(url, headers={'User-Agent': 'unsloth-buddy/1.0'})

req = urllib.request.Request(url, headers={'User-Agent': 'unsloth-buddy/1.0'})

with urllib.request.urlopen(req) as resp:

def fetch_text(url):     req = urllib.request.Request(url, headers={'User-Agent': 'unsloth-buddy/1.0'})

req = urllib.request.Request(url, headers={'User-Agent': 'unsloth-buddy/1.0'})

with urllib.request.urlopen(req) as resp:

manifest_url = "https://unpkg.com/getdesign@latest/templates/manifest.json"

if args.fetch:         print(f"\n--- DESIGN.md for {match['brand']} ---\n")         design_url = f"https://unpkg.com/getdesign@latest/templates/{match['file']}"

print(f"\n--- DESIGN.md for {match['brand']} ---\n")         design_url = f"https://unpkg.com/getdesign@latest/templates/{match['file']}"         try:

design_url = f"https://unpkg.com/getdesign@latest/templates/{match['file']}"

Executable file

exec(open("setup_colab.py").read())

ok2, out2 = _run(             "wget -qO- https://raw.githubusercontent.com/unslothai/unsloth/main/unsloth/_auto_install.py | python -"

"wget -qO- https://raw.githubusercontent.com/unslothai/unsloth/main/unsloth/_auto_install.py | python -"

"wget -qO- https://raw.githubusercontent.com/unslothai/unsloth/main/unsloth/_auto_install.py | python -"

Executable file

def fetch_metrics(port=8080):     try:         r = requests.get(f"http://localhost:{port}/api/metrics", timeout=2)

try:         r = requests.get(f"http://localhost:{port}/api/metrics", timeout=2)         r.raise_for_status()

r = requests.get(f"http://localhost:{port}/api/metrics", timeout=2)

print(f"Connecting to Gaslamp training process on http://localhost:{args.port}...")

Executable file

Executable file

Executable file

Executable file

# {"messages": [ #     {"role": "user", "content": [ #         {"type": "image", "image": "https://..."},

#     {"role": "user", "content": [ #         {"type": "image", "image": "https://..."}, #         {"type": "text", "text": "Describe this image."}

#         {"type": "image", "image": "https://..."},

# 5. Train with Gaslamp Dashboard # ============================================================ # Web dashboard available at http://localhost:8080/

# Web dashboard available at http://localhost:8080/

Executable file

Executable file

# {"messages": [ #     {"role": "user", "content": [ #         {"type": "image", "image": "https://..."},

#     {"role": "user", "content": [ #         {"type": "image", "image": "https://..."}, #         {"type": "text", "text": "Describe this image."}

#         {"type": "image", "image": "https://..."},

@import url('https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600&family=JetBrains+Mono:wght@400;500&display=swap');

function getBaseUrl() { return `http://localhost:${portSlider.value}`; }

async function checkConnection() {   try {     const r = await fetch(`${getBaseUrl()}/v1/models`, { signal: AbortSignal.timeout(3000) });

try {     const r = await fetch(`${getBaseUrl()}/v1/models`, { signal: AbortSignal.timeout(3000) });     if (r.ok) {

const r = await fetch(`${getBaseUrl()}/v1/models`, { signal: AbortSignal.timeout(3000) });

try {     const resp = await fetch(`${getBaseUrl()}/v1/chat/completions`, {

const resp = await fetch(`${getBaseUrl()}/v1/chat/completions`, {

<meta name="viewport" content="width=device-width, initial-scale=1.0">   <title>Gaslamp | Training Dashboard</title>   <script src="https://cdn.tailwindcss.com"></script>

<title>Gaslamp | Training Dashboard</title>   <script src="https://cdn.tailwindcss.com"></script>   <script src="https://cdn.jsdelivr.net/npm/chart.js"></script>

<script src="https://cdn.tailwindcss.com"></script>

<script src="https://cdn.jsdelivr.net/npm/chart.js"></script>

@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap');

<div class="flex items-center space-x-4">         <div class="w-14 h-14 rounded-xl bg-gray-900 border border-gray-700 shadow-lg shadow-emerald-500/20 flex items-center justify-center p-1">           <svg viewBox="0 0 64 64" fill="none" xmln

<div class="w-14 h-14 rounded-xl bg-gray-900 border border-gray-700 shadow-lg shadow-emerald-500/20 flex items-center justify-center p-1">           <svg viewBox="0 0 64 64" fill="none" xmlns="http://www.w3.org/2000/svg" class="w-10 h-10"> 

<svg viewBox="0 0 64 64" fill="none" xmlns="http://www.w3.org/2000/svg" class="w-10 h-10">

<footer class="text-center text-xs text-gray-600 pb-4">     Gaslamp Training Dashboard · Powered by Unsloth-buddy ·     <a href="https://github.com/TYH-labs/unsloth-buddy" class="underline hover:text-gray-400">GitHub</a>

Gaslamp Training Dashboard · Powered by Unsloth-buddy ·     <a href="https://github.com/TYH-labs/unsloth-buddy" class="underline hover:text-gray-400">GitHub</a>   </footer>

<a href="https://github.com/TYH-labs/unsloth-buddy" class="underline hover:text-gray-400">GitHub</a>

pollTimer = setInterval(async () => {     try {       const r = await fetch(`http://localhost:${port}/api/metrics`);

try {       const r = await fetch(`http://localhost:${port}/api/metrics`);       if (r.ok) {

const r = await fetch(`http://localhost:${port}/api/metrics`);

function connect(port=8080) {   const connBadge = document.getElementById('conn-badge');   const es = new EventSource(`http://localhost:${port}/api/stream`);

const es = new EventSource(`http://localhost:${port}/api/stream`);

es.onopen = () => {     // Also do an immediate HTTP fetch to get the full snapshot (handles reconnection)     fetch(`http://localhost:${port}/api/metrics`)

// Also do an immediate HTTP fetch to get the full snapshot (handles reconnection)     fetch(`http://localhost:${port}/api/metrics`)       .then(r => r.ok ? r.json() : null)

fetch(`http://localhost:${port}/api/metrics`)

// Detect port from URL hash, e.g. http://localhost:8080/#port=9090

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

@import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@300;400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap');

@import url('https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@300;400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%238a8a8a' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%238a8a8a' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

@import url('https://fonts.googleapis.com/css2?family=DM+Sans:ital,opsz,wght@0,9..40,300;0,9..40,400;0,9..40,500;0,9..40,600;1,9..40,400&family=Instrument+Serif:ital@0;1&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%236b7280' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

<style>     /* === FONT IMPORT === */     @import url('https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@300;400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap');

@import url('https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@300;400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap');

transition: border-color 0.2s ease, box-shadow 0.2s ease;         appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5

appearance: none;         background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%238a8a8a' stroke-width='1.5' fill='none' stroke-linec

background-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='12' height='8' viewBox='0 0 12 8'%3E%3Cpath d='M1 1l5 5 5-5' stroke='%238a8a8a' stroke-width='1.5' fill='none' stroke-linecap='round'/%3E%3C/svg%3E")

<footer>             Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

Built with <a href="https://github.com/TYH-labs/unsloth-buddy" target="_blank">unsloth-buddy</a> · <a href="https://gaslamp.dev" target="_blank">Gaslamp 🔥</a>

# gaslamp.md — Roadbook for `{project_name}` # Powered by Gaslamp · https://gaslamp.dev

# Powered by Gaslamp · https://gaslamp.dev