Lambent/cosmo-upscale
Partial fine-tune vs HuggingFaceTB/cosmo-1b. Partial change: 4% of the weights changed (attention, mlp, norms), in 1 of 24 layers.
Tensors changed: Embeddings 0/1, Norms 2/49, MLP 3/72, Attention 4/96, lm_head 0/1
Partial fine-tune vs HuggingFaceTB/cosmo-1b. Partial change: 4% of the weights changed (attention, mlp, norms), in 1 of 24 layers.
Tensors changed: Embeddings 0/1, Norms 2/49, MLP 3/72, Attention 4/96, lm_head 0/1