AstralX 3B
AstralX is a lightweight, specialized code-assistant model fine-tuned on top of Qwen2.5-Coder-3B-Instruct.
It is designed to act as an efficient, low-latency secondary AI copilot directly in your local development environment or CLI setup. While it might not outsmart massive frontier models on complex architecture design, it serves as an excellent, fast companion for everyday coding tasks, refactoring, and autocomplete.
🚀 Model Details
- Base Model: Qwen2.5-Coder-3B-Instruct
- Parameters: 3 Billion
- Fine-tuning Dataset: 6,000+ curated, high-quality code samples extracted and filtered from top-tier frontier models (Claude 3.5 Sonnet, Gemini 1.5 Pro, DeepSeek-V3, Qwen-Max, GLM-4).
- Primary Focus: High-quality syntax, idiomatic patterns, and optimal formatting.
📚 Supported Languages
The dataset was specially curated to boost performance in the following languages:
- 🦀 Rust (Idiomatic memory management, ownership patterns, and CLI tools)
- 🐍 Python (Scripts, automation, backend logic)
- 🌙 Lua (Game scripting, Neovim configs, embeds)
- ☕ Java (Object-oriented architecture, clean code templates)
- 🌐 JavaScript (Modern ES6+, web tools)
💡 Use Case & Philosophy
AstralX doesn't aim to be a "know-it-all" general intelligence. It was built with the "vibe-coding" philosophy in mind—providing high-speed, localized assistance. It works perfectly as an auxiliary tool running locally via Ollama or embedded directly into text editors.
🛠️ How to Use with Ollama (GGUF)
If you are using the GGUF version with the provided Modelfile, you can run it locally via:
1ollama create AstralX -f ./Modelfile
2ollama run AstralX
Fine-tuned with love for high-speed terminal and editor workflows.
