-
-
-
-
modular-transformer60b361ff · ·
New modular shape-transformer architecture. Added Embeddings and Generative Head as own modules.
-
Vanilla-Transformer34f876c3 · ·
Full Attention Vanilla Transformer with: num_positions: 4096 num_layers: 8 embed_dim: 64 num_heads: 4 running at 10,5 GB