gptq▌

Post-training quantization method that compresses LLMs to 4-bit with minimal accuracy loss using group-wise quantization.