Qwen Reasoning Trainer

About

A Go implementation for abliteration, reasoning-trace fine-tuning, and inference of the Qwen3.8 27B model in Q4_K_M GGUF quantization, with a CPU backend and a Vulkan GPU backend. It trains the model on reasoning traces exported from a teacher model's coding sessions, aiming to make this feasible on consumer hardware. It also builds an OpenAI-compatible inference server.