mgpc 的提交内容

How much LLM training data is there, in the limit?

Revised Chinchilla scaling implies higher compute requirements for small models