A more memory-efficient rewrite of the HF transformers implementation of Llama for use with quantized weights.
By chatting or signing in you agree to the Terms and chat-message logging (revocable in History).