Search papers, labs, and topics across Lattice.
3
0
4
Time series language models can achieve up to 7.68脳 faster inference and improved performance by intelligently compressing tokens based on their information structure.
Channel-wise adaptive learning rates in Gated Delta Networks unlock superior long-context recall, rivaling softmax attention without the quadratic cost.
By strategically amplifying updates along flat directions in the loss landscape, LITE unlocks faster LLM pre-training with existing matrix-based optimizers like Muon and SOAP.
