Search papers, labs, and topics across Lattice.
2
0
3
Pushing KV cache quantization down to 2 bits without sacrificing downstream accuracy is now viable by rotating key and value matrices according to their joint second-order interaction statistics.
Legal agents frequently reach the correct legal outcome through entirely fabricated citations and broken tool-use trajectories, revealing a deceptive "Right-Answer-Wrong-Reason" failure mode invisible to standard benchmarks.
