XL-LuaCopilot-0.6B-FFT-MNN_Q8

XL-LuaCopilot-0.6B-FFT

XL-LuaCopilot-0.6B-FFT is a large language model (LLM) based on the Qwen architecture(Qwen3-0.6B-Base), specifically designed for code generation tasks in Lua programming language. It has been full fine-tuned (FFT) to improve its performance and efficiency when generating Lua code.

I sugggest you use "chat_template_kwargs": {"enable_thinking": false} because my train data with none thinking. I also found low temperature ususually works well for code generation tasks.

How To Use

I'm trying to use MNN (faster than llama.cpp), but the documentation is confusing. I followed the docs to generate the MNN model weights, but their usage is still unknown.

Train Device

Online GPU is Expensive !