Closed binganao closed 1 year ago
Great job! Here is a question, how did the team extend the sequence length of llama from 2048 to 4096?
They just trained in on 4096 tokens lol. It's a variable that can be selected while training, wasn't any magic.
Great job! Here is a question, how did the team extend the sequence length of llama from 2048 to 4096?