Fox-AI-by-teolm30/Fox-1.5-Nova
teolm30/Fox-1.5-Nova is a 7 billion parameter Qwen2-based model fine-tuned by teolm30, optimized for coding, reasoning, and general assistance. Designed for fast local inference with full FP16 precision, it offers strong performance in code generation and instruction following. This model is particularly suited for developers seeking a high-speed, locally deployable coding assistant on consumer GPUs.
Loading preview...
Fox 1.5 Nova: Fast Local Inference for Coding and Reasoning
Fox 1.5 Nova is a 7 billion parameter model built on the Qwen2 architecture, fine-tuned by teolm30. It is specifically optimized for coding, reasoning, and general assistance, designed to run efficiently with full FP16 precision on consumer-grade GPUs.
Key Capabilities & Performance
- Optimized for Speed: Achieves approximately 40+ tokens/second on typical generation lengths on RTX 3090/4090 GPUs, making it ideal for fast local development workflows.
- Strong Coding Performance: Scores 67.4 on HumanEval for code generation and 74.8 on GSM8K for grade-school math, demonstrating solid capabilities in programming and arithmetic reasoning.
- Instruction Following: Excels in multi-turn conversations with an MT-Bench score of 8.1, indicating good instruction adherence.
- Local Deployment: Requires around 14GB VRAM for FP16, making it accessible for local inference without quantization.
Good For
- Local Coding Assistance: Developers needing a fast, private, and locally deployable AI for code generation and debugging.
- Privacy-Sensitive Deployments: Ideal for scenarios where data must remain on-premises.
- General Development Workflows: Enhancing productivity with quick responses for reasoning and general assistance tasks.
While it trades raw intelligence in expert-level reasoning (e.g., GPQA, MATH) and multimodal capabilities compared to frontier models like Opus 4.6, Fox 1.5 Nova provides a compelling balance of performance and local deployability.