catid/cat-llama-3-70b-awq-q256-w4-gemm

AI Model Name: Llama 3 70B "Built with Meta Llama 3" https://llama.meta.com/llama3/license/

This is the result of running AutoAWQ to quantize the LLaMA-3 70B model to ~4 bits/parameter.

To launch an OpenAI-compatible API endpoint on your Linux server with 2x 3090 or 4090 GPUs:

git lfs install
git clone https://huggingface.co/catid/cat-llama-3-70b-awq-q256-w4-gemm

conda create -n vllm70 python=3.10 -y && conda activate vllm70

pip install -U git+https://github.com/vllm-project/vllm.git

python -m vllm.entrypoints.openai.api_server --model cat-llama-3-70b-awq-q256-w4-gemm --tensor-parallel-size 2 --gpu-memory-utilization 0.935

Sadly this barely doesn't fit by ~300MB or so.