sashakunitsyn/vlrm-blip2-opt-2.7b
Partial fine-tune vs Salesforce/blip2-opt-2.7b. Partial change: 3% of the weights changed (attention, norms, other), in 12 of 39 layers.
Tensors changed: Embeddings 0/5, Norms 62/350, MLP 0/284, Attention 144/556, other 51/51