Search papers, labs, and topics across Lattice.
2
0
3
1
Key contribution not extracted.
Stop wasting compute: PonderLM-3 learns to spend extra inference FLOPs only on the tokens that actually need them, outperforming fixed-step pondering methods.
