ai4bharat/Cadence-Fast
Partial fine-tune vs google/gemma-3-270m. Partial change: 37% of the weights changed (attention, mlp, norms), in 18 of 18 layers.
Tensors changed: Embeddings 0/1, Norms 71/109, MLP 54/54, Attention 72/72
Partial fine-tune vs google/gemma-3-270m. Partial change: 37% of the weights changed (attention, mlp, norms), in 18 of 18 layers.
Tensors changed: Embeddings 0/1, Norms 71/109, MLP 54/54, Attention 72/72