Search papers, labs, and topics across Lattice.
Affiliation:
3
0
6
TailSieve achieves up to 2.59x speedup in LLM rollouts by intelligently routing long-tail requests, transforming how we handle high-concurrency decoding.
A unified end-to-end training framework for TTS systems achieves a groundbreaking 0.78% word error rate, setting a new standard in the field.
Forget random noise – teaching models *how* to explore their reasoning process yields more reliable inference-time scaling.
