3 ms·
Exactly. It's inefficient for LLMs to recompute state based on previous output, but that's what we have so far, and it works, surprisingly.
by ukuina 3y ago
Exactly. It's inefficient for LLMs to recompute state based on previous output, but that's what we have so far, and it works, surprisingly.