nvidia/Nemotron-3-Super-120B-A12B-BF16-MTPv2
Full fine-tune (fewer layers) vs nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16. nvidia/Nemotron-3-Super-120B-A12B-BF16-MTPv2 matches part of nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 (1040 of its 42683 tensors; the rest of nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 2 of 2 layers.
Tensors changed: other 3/3, Norms 5/5, Attention 4/4, MLP 70/70