The easiest way is to consider a token a word. That isnt exact but it is enough of a framework.
To give you an idea, Llama3 was trained on 16 trillion tokens.
You are viewing a single comment's thread from:
The easiest way is to consider a token a word. That isnt exact but it is enough of a framework.
To give you an idea, Llama3 was trained on 16 trillion tokens.
What I can understand we have to feed a lots of data..
That is the baseline if you want to go from that.