Search papers, labs, and topics across Lattice.
1
0
1
Gated attention's promise of eliminating attention sinks evaporates at million-token scale, proving that sink tokens and massive activations are fundamentally driven by the training objective rather than the attention architecture.