fbolzan/BZN-LLM-v1
fbolzan/BZN-LLM-v1 is a 7.6 billion parameter language model developed by Fernando Bolzan, based on Qwen2.5-Coder-7B-Instruct. Specialized in code generation and reasoning, it was fine-tuned using QLoRA on an RTX 3060 12GB. This model is optimized for tasks requiring strong logical deduction and programming capabilities, supporting both Portuguese and English.
Loading preview...
BZN-LLM-v1 Overview
BZN-LLM-v1 is a 7.6 billion parameter language model developed by Fernando Bolzan, specifically designed for code and reasoning tasks. It is built upon the Qwen2.5-Coder-7B-Instruct base model and was fine-tuned using the QLoRA method on an RTX 3060 12GB. While specific internal architecture and detailed training data are not publicly disclosed, its specialization targets robust performance in programming and logical problem-solving.
Key Capabilities
- Code Generation: Optimized for generating and understanding code.
- Reasoning: Enhanced for tasks requiring logical deduction.
- Multilingual Support: Supports both Portuguese and English.
- Hardware Accessibility: Fine-tuned on consumer-grade hardware, suggesting potential for efficient deployment.
Formats and Deployment
The model is available in standard safetensors format for Transformers and in various GGUF quantizations for efficient use with tools like Ollama (version 0.30+). Available GGUF formats include bf16 (15.2 GB for maximum quality), q5_k_m (5.2 GB for high quality and lighter footprint), and q4_k_m (4.5 GB for good cost-benefit).
Good For
- Developers needing a specialized model for code-related tasks.
- Applications requiring strong reasoning abilities.
- Users looking for a model with multilingual support (PT/EN) in a technical context.
- Deployment on systems with limited VRAM using quantized GGUF versions.