A more memory-efficient rewrite of the HF transformers implementation of Llama for use with quantized weights. - View it on GitHub