Quant of https://huggingface.co/junelee/wizard-vicuna-13b tested working with Occam's KoboldAI/GPTQ.
Someone made a Triton quant already here, but it will not work with Occam's KoboldAI/GPTQ fork: https://huggingface.co/fbjr/wizard-vicuna-13b-4bit-128g
Note that this model is fairly heavily censored (in my opinion) and delivers AI-moralizing responses to prompts that Vicuna 1.1 does not complain about.
python llama.py ./wizard-vicuna-13b c4 --wbits 4 --true-sequential --groupsize 128 --save_safetensors 4bit-128g.safetensors
- Downloads last month
- 12
This model does not have enough activity to be deployed to Inference API (serverless) yet.
Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.