Avara X1 Mini fine-tunes Qwen2.5 for edge coding

A hobbyist fine-tune of an already capable small model — interesting for edge deployments but light on evidence it actually outperforms the base Qwen2.5-1.5B.

• –Based on Qwen2.5-1.5B (titled as "2B" — slightly inflated framing)
• –Fine-tuned using Unsloth on The Stack (BigCode), Open-Platypus, and math competition datasets
• –Provides Q4_K_M GGUF for quantized local inference and a LoRA adapter for further customization
• –No published benchmarks to substantiate the "powerhouse" claim — community will need to self-evaluate
• –Apache 2.0 license makes it freely usable and modifiable; Discord community available for feedback