Open-weight 7B model that surpasses ChatGPT-3.5 on benchmarks while running on consumer hardware. C-RLFT fine-tuned on mixed-quality data — punches far above its weight.