Search papers, labs, and topics across Lattice.
Affiliation:
1
0
Placing reasoning traces before context in transformers can boost long-context reasoning accuracy by up to 100% while using exponentially less memory.
