TIDE

Python ★ 32 updated 3mo ago

Dynamic per-token early exit for LLM inference. Skip layers tokens don't need

No plain-English explanation yet — one is being written right now. Check back in a minute.