syvai/hviske-v5.3

Full fine-tune vs syvai/hviske-v5.1. Fine-tune: 100% of the weights changed, across 48 of 48 layers.

Tensors changed: Norms 321/770, other 462/462, MLP 386/386, Attention 528/528, Embeddings 2/4

On the Hub · Downloads on Model Pulse