Unweight: how we compressed an LLM 22% without sacrificing quality

https://blog.cloudflare.com/unweight-tensor-compression/