Search papers, labs, and topics across Lattice.
1
0
3
Not all attention heads are created equal: selectively aggregating the *right* ones dramatically sharpens the interpretability of text-to-image diffusion models.