Ameya-Kawade/cmdcaliper

Full fine-tune vs thenlper/gte-base. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: Embeddings 3/3, Attention 96/96, other 50/50