Brewmode Qwen3-8B v3
Fine-tuned on 100K high-quality code examples (Magicoder + OpenCodeInstruct). Trained with Unsloth on H100, 1000 steps, final loss 0.699.
Training Data
- Magicoder-OSS-Instruct-75K (open-source inspired code)
- Magicoder-Evol-Instruct-110K (evolved complexity)
- OpenCodeInstruct top 30K (test-verified, score >= 0.9)
- Filtered for web/code relevance (HTML, JS, TS, React, API)
Usage
Use with llama.cpp, Ollama, or any GGUF-compatible runtime.
- Downloads last month
- 12
Hardware compatibility
Log In to add your hardware
4-bit
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support