6 ms·

What do you mean?

• hiccuphippo 7mo ago
  The article says the LLM has to load 15540 tokens every time, I wonder if that can be reduced while retaining the context maybe with deduplications, removing superfluous words, using shorter expressions with the same meaning or things like that.