Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
11
Reasoning supervision can be compressed over 450x with uncertainty-guided contraction, boosting LLM efficiency without sacrificing capability.
