grimjim/Magnolia-v3-medis-dilute-12B

Partial fine-tune vs grimjim/Magnolia-v3-12B. Partial change: 12% of the weights changed (attention, embeddings, mlp, norms), in 22 of 40 layers.

Tensors changed: lm_head 0/1, Embeddings 1/1, Norms 6/81, MLP 9/120, Attention 10/160

2 derivatives checked

On the Hub · Downloads on Model Pulse