mnoukhov/nuevamol-46m-reinvent-sft
Full fine-tune vs mnoukhov/nuevamol-46M-6Btok-wd0.5. Fine-tune: 100% of the weights changed, across 13 of 13 layers.
Tensors changed: Embeddings 1/1, Norms 21/27, MLP 39/39, Attention 52/52
Full fine-tune vs mnoukhov/nuevamol-46M-6Btok-wd0.5. Fine-tune: 100% of the weights changed, across 13 of 13 layers.
Tensors changed: Embeddings 1/1, Norms 21/27, MLP 39/39, Attention 52/52