rhymes-ai/Aria-Base-64K

Full fine-tune vs rhymes-ai/Aria-Base-8K. Fine-tune: 98% of the weights changed, across 28 of 28 layers.

Tensors changed: Embeddings 1/4, Norms 60/171, MLP 170/278, Attention 121/337, other 1/1, lm_head 1/1

1 derivatives checked

On the Hub · Downloads on Model Pulse