Lambent/Mira-v1.28-wave

Full fine-tune vs Lambent/Mira-v1.25-27B-Wave. Fine-tune: 100% of the weights changed, across 62 of 62 layers.

Tensors changed: Embeddings 2/4, Norms 349/484, MLP 261/294, Attention 405/464, other 1/1

3 derivatives checked

On the Hub · Downloads on Model Pulse