teolm30/Fox-1.5-Nova

Fox 1.5 Nova: Code Generation Model

Fox 1.5 Nova is a 7 billion parameter code generation model developed by teolm30, fine-tuned from DeepSeek-Coder-7B-Instruct. It is specifically optimized for competitive programming, systems design, and generating real-world code patterns across more than 50 programming languages.

Key Capabilities & Features

• Specialized Code Generation: Excels in competitive programming scenarios, systems design, and practical coding tasks.
• Broad Language Support: Fine-tuned to handle code patterns across 50+ programming languages.
• Efficient Local Inference: Designed for local deployment, offering approximately 40+ tokens/second inference speed on fp16.
• Low Resource Footprint: Requires only about 6GB VRAM for the 4-bit quantized version, making it accessible for consumer-grade hardware.
• Cost-Effective: Free to use, eliminating API costs associated with larger, proprietary models.
• Offline Operation: Does not require an internet connection for inference.

Performance & Technical Details

Built using QLoRA (4-bit NF4) with a LoRA r of 16 and alpha of 64, the model underwent 220 training steps over 10 epochs. It has a maximum output length of 512 tokens. While it does not have built-in tool-use capabilities, it can be integrated with external frameworks like OpenClaw for agentic workflows.

Ideal Use Cases

• Local Code Generation: Developers needing a powerful code assistant that runs entirely on their machine.
• Competitive Programming: Generating solutions or boilerplate for programming contests.
• Systems Development: Assisting with code for system-level applications and design.
• Educational Purposes: Learning and experimenting with code generation without incurring API costs.

This model provides a robust, resource-efficient solution for a wide range of code-centric applications, particularly where local execution and specific code pattern generation are critical.