Search papers, labs, and topics across Lattice.
Affiliation:
1
0
3
Transformer weight-space learning is now invariant to the full, rich symmetry group of attention heads and token embeddings.
