Can we reduce the LLM model size during the training?

Instead of compressing a model after training, the paper proposes compressing during training by using control theory (balanced truncation… Continue reading on Medium »