sergeyzh/rubert-mini-frida
Full fine-tune vs sergeyzh/rubert-mini-sts. Fine-tune: 98% of the weights changed, across 7 of 7 layers.
Tensors changed: Embeddings 4/5, Norms 28/28, Attention 56/56, other 28/30
Full fine-tune vs sergeyzh/rubert-mini-sts. Fine-tune: 98% of the weights changed, across 7 of 7 layers.
Tensors changed: Embeddings 4/5, Norms 28/28, Attention 56/56, other 28/30