The latest articles on Forem by Bowen (@tuwenbo). https://forem.com/tuwenbo https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3770846%2Fbe2cd2c3-6413-4344-94ef-d590f4ffa168.jpg https://forem.com/tuwenbo en Bowen Fri, 13 Feb 2026 11:52:01 +0000 https://forem.com/tuwenbo/i-built-a-zero-code-llm-fine-tuning-app-for-mac-heres-how-it-works-4027 https://forem.com/tuwenbo/i-built-a-zero-code-llm-fine-tuning-app-for-mac-heres-how-it-works-4027 <h2> The Problem </h2> <p>Fine-tuning LLMs locally on a Mac usually means:</p> <ul> <li>Wrestling with Python environments</li> <li>Writing training scripts</li> <li>Managing model weights manually</li> <li>Figuring out how to actually <em>use</em> your fine-tuned model</li> </ul> <p>I wanted something simpler.</p> <h2> The Solution: M-Courtyard </h2> <p>M-Courtyard is an open-source macOS desktop app that handles the entire <br> fine-tuning pipeline through a visual interface:</p> <ol> <li> <strong>Import</strong> your documents (txt, pdf, docx)</li> <li> <strong>Generate</strong> training data automatically using your local Ollama model</li> <li> <strong>Fine-tune</strong> with LoRA — watch the loss curve in real-time</li> <li> <strong>Test</strong> your model with built-in chat</li> <li> <strong>Export</strong> to Ollama with one click (Q4/Q8/F16 quantization)</li> </ol> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F3mir80h5ufzis93fqq1l.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F3mir80h5ufzis93fqq1l.png" alt=" " width="800" height="508"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fne3wl6rh9prrkpf8cujd.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fne3wl6rh9prrkpf8cujd.png" alt=" " width="800" height="508"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Ff9apvkz458v720dhgd87.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Ff9apvkz458v720dhgd87.png" alt=" " width="800" height="508"></a></p> <h3> Everything Runs Locally </h3> <p>No cloud services. No API keys. No data leaves your Mac. </p> <p>It's built on Apple's MLX framework, which takes full advantage of <br> Apple Silicon's unified memory architecture.</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fmouejwh2lkc72z9xi9cj.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fmouejwh2lkc72z9xi9cj.png" alt=" " width="800" height="508"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F2gjm877761refsl9krrh.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F2gjm877761refsl9krrh.png" alt=" " width="800" height="508"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fha6q2t87bwwyobtpziml.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fha6q2t87bwwyobtpziml.png" alt=" " width="800" height="508"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F9de2imdb8v1bxv3pqnlu.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F9de2imdb8v1bxv3pqnlu.png" alt=" " width="800" height="508"></a></p> <h3> Supported Models </h3> <p>Works with models from the mlx-community on Hugging Face:</p> <ul> <li>Qwen 3 (0.6B to 14B)</li> <li>DeepSeek R1</li> <li>GLM 5</li> <li>Llama 3</li> <li>And many more</li> </ul> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F72bh5q67x9s2n1g8s6eg.gif" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F72bh5q67x9s2n1g8s6eg.gif" alt=" " width="720" height="450"></a></p> <h3> One-Click Export to Ollama </h3> <p>After fine-tuning, export your model directly to Ollama. <br> Then just run <code>ollama run your-model-name</code>.</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fu0wndq2r1pmbpfxk8vhb.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fu0wndq2r1pmbpfxk8vhb.png" alt=" " width="800" height="508"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F6cqyc1oia4ocwaweanl8.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F6cqyc1oia4ocwaweanl8.png" alt=" " width="800" height="508"></a></p> <h2> Tech Stack </h2> <ul> <li> <strong>Frontend</strong>: React + TypeScript</li> <li> <strong>Backend</strong>: Rust (Tauri 2.x)</li> <li> <strong>ML Engine</strong>: mlx-lm (Apple MLX)</li> <li> <strong>Model Runtime</strong>: Ollama</li> <li> <strong>License</strong>: AGPL 3.0</li> </ul> <h2> Try It Out </h2> <ul> <li> <strong>GitHub</strong>: <a href="https://github.com/Mcourtyard/m-courtyard" rel="noopener noreferrer">github.com/tuwenbo0120/m-courtyard</a> </li> <li> <strong>Download</strong>: <a href="https://github.com/Mcourtyard/m-courtyard/releases/latest" rel="noopener noreferrer">Latest Release (.dmg)</a> </li> <li> <strong>Discord</strong>: <a href="https://discord.gg/hjkrHWrQ" rel="noopener noreferrer">Join the community</a> </li> </ul> <p>Requirements: macOS 14+, Apple Silicon (M1/M2/M3/M4), 16GB+ RAM recommended</p> <p>Feedback and contributions welcome! ⭐</p> xamarin opensource machinelearning ai