tsumeone's picture
Create README.md
53c85a7

Quant of https://huggingface.co/chavinlo/gpt4-x-alpaca

There's already one located at https://huggingface.co/anon8231489123/gpt4-x-alpaca-13b-native-4bit-128g, but neither the triton nor cuda version they uploaded seem to want to work on older versions of GPTQ-for-LLaMA such as the one currently used with KoboldAI for 4bit support on 0cc4m's fork.

This was quantized with cuda, not triton.

python llama.py ./gpt4-x-alpaca c4 --wbits 4 --true-sequential --groupsize 128 --save_safetensors gpt-x-alpaca-13b-native-4bit-128g-cuda.safetensors