francesca9805/zho-100mb-after-wc-uniform-newlex-after-ckpt500-packed-bfdiso_seed455
Full fine-tune vs francesca9805/ppt-wc-uniform-newlex-zho-after-100mb-packed-bfdiso_seed455. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Attention 48/48, Norms 25/50, MLP 48/48, Embeddings 1/1