3 ms·

This isn’t talking about compaction. This refers to performance as the model is loaded with 500k to 1m tokens.

• radicality 6mo ago
  Ah, thanks, makes sense, I’ll read more about this