Tag: learned-representation

papers

Neural Weight Compression

Paper note on using a single learned neural codec to compress whole LLM-scale weight sets instead of relying only on handcrafted quantization formats.