tsumeone/vicuna-13B-1.1-GPTQ-4bit-128g-cuda

Quant of https://huggingface.co/TheBloke/vicuna-13B-1.1-HF

There's already one located at https://huggingface.co/TheBloke/vicuna-13B-1.1-GPTQ-4bit-128g, but neither version they uploaded works with certain older versions of GPTQ-for-LLaMA (such as 0cc4m's fork that is used with their fork of KoboldAI).

This was quantized with 0cc4m's fork of GPTQ-for-LLaMA.

python llama.py ./vicuna-13B-1.1-HF c4 --wbits 4 --true-sequential --groupsize 128 --save_safetensors 4bit-128g.safetensors