What Changes When We Compress a Language Model?

I wanted a smaller, faster model. Instead, I discovered a behavior I wasn’t expecting Continue reading on Medium »