--- tags: - gguf - llama.cpp - unsloth - vision-language-model - qwen - typescript license: mit datasets: - mhhmm/typescript-instruct-20k base_model: - Qwen/Qwen3.5-4B --- # Qwen3.5-4B-TypeScript-Coder : GGUF This model is a high-performance fine-tune of **Qwen 3.5 4B**, specifically optimized for **TypeScript development**, architectural reasoning, and full-stack engineering. Fine-tuned using **Unsloth Studio**, it leverages Qwen 3.5's native multimodal foundation to provide industry-leading code generation and visual-to-code capabilities. ## 🚀 Key Features * **TypeScript Specialization:** Deeply tuned for strict type safety, Generics, and modern frameworks like React, Next.js, and Node.js. * **Visual-to-Code:** Capable of understanding UI screenshots and system diagrams to generate clean, type-safe logic. * **Optimized Inference:** Converted to GGUF for low-latency performance on local hardware. ## 🤝 Dataset Credits This model was trained using the **[typescript-instruct-20k](https://huggingface.co/datasets/mhhmm/typescript-instruct-20k)** dataset by **mhhmm**. This high-quality data allows the model to handle everything from simple scripts to enterprise-level refactoring. ## 📂 Model Files & Inference Compatible with `llama.cpp` and other GGUF-supported runners. * **High-Precision:** `qwen3.5-4b-typescript.Q8_0.gguf` * **Vision Projector:** `qwen3.5-4b-typescript.BF16-mmproj.gguf` **Example usage**: * **CLI Chat:** `llama-cli -hf MassivDash/qwen3.5-4B-typescript-coder --jinja` * **Vision Tasks:** `llama-mtmd-cli -hf MassivDash/qwen3.5-4B-typescript-coder --jinja` ## ⚠️ Ollama Integration To use this multimodal model in Ollama: 1. Create a `Modelfile` in your local directory. 2. Run: `ollama create qwen-ts-coder -f ./Modelfile` ## 🔗 Resources * **Author Blog:** Find more tutorials at [spaceout.pl](https://spaceout.pl) * **Training:** This model was trained **2x faster** with [Unsloth](https://github.com/unslothai/unsloth). [](https://github.com/unslothai/unsloth)