NeuML/colbert-bert-tiny
Partial fine-tune vs google/bert_uncased_L-2_H-128_A-2. Partial change: 83% of the weights changed (attention, embeddings, norms, other), in 2 of 2 layers.
Tensors changed: Embeddings 3/4, Norms 8/8, Attention 16/16, other 8/10