fpadovani/eng-100mb-after-wc-zipf-newlex-jpn-ckpt4000_seed3407_seed3407
Full fine-tune vs fpadovani/ppt-wc-zipf-newlex-jpn-100mb_seed3407. 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