SqueezeLLM: Dense-and-Sparse Quantization

I dunno why this was passed by HN. Those results are fantastic.