nvidia/Nemotron-H-4B-Instruct-128K
Full fine-tune vs nvidia/Nemotron-H-4B-Base-8K. Fine-tune: 100% of the weights changed, across 52 of 52 layers.
Tensors changed: Embeddings 1/1, other 167/168, Norms 77/77, MLP 48/48, Attention 16/16, lm_head 1/1
Full fine-tune vs nvidia/Nemotron-H-4B-Base-8K. Fine-tune: 100% of the weights changed, across 52 of 52 layers.
Tensors changed: Embeddings 1/1, other 167/168, Norms 77/77, MLP 48/48, Attention 16/16, lm_head 1/1