A arxiv.org Lossless LLM compression for efficient GPU inference via dynamic-length float 411 points CharlesW 1 year ago 117 comments
