hotchpotch/bekko-embedding-v1-a8m

Full fine-tune vs hotchpotch/bekko-embedding-v1-a8m-pt. Fine-tune: 100% of the weights changed, across 4 of 4 layers.

Tensors changed: Embeddings 2/2, Norms 8/8, Attention 8/8, MLP 8/8

1 derivatives checked