ai4privacy/llama-ai4privacy-multilingual-categorical-anonymiser-openpii
Partial fine-tune vs answerdotai/ModernBERT-base. Partial change: 74% of the weights changed (attention, embeddings, mlp, norms, other), in 22 of 22 layers.
Tensors changed: other 1/1, Norms 45/45, Embeddings 1/2, Attention 44/44, MLP 44/44
1 other repos hold byte-for-byte identical weight files.