What you want is to either use a supercot Lora, or to use a finetuned model from HF with SuperCOT in its dataset: https://huggingface.co/kaiokendev/SuperCOT-LoR

What you want is to either use a supercot Lora, or to use a finetuned model from HF with SuperCOT in its dataset: https://huggingface.co/kaiokendev/SuperCOT-LoRA https://huggingface.co/kaiokendev/SuperCOT-LoRA https://huggingface.co/search/full-text?q=Supercot&type=model https://huggingface.co/search/full-text?q=Supercot&type=mode... As an aside, I would suggest using a llama.cpp Q5_1 model now. Its more accurate, and with CUDA acceleration, its basically as fast as gptq. You also have the option of running models that dont quite fit into vram, with a very small performance hit.