nex-agi/Nex-N2.5-Max-DSpark
Full fine-tune (fewer layers) vs nex-agi/Nex-N2.5-Max. nex-agi/Nex-N2.5-Max-DSpark matches part of nex-agi/Nex-N2.5-Max (3517 of its 143438 tensors; the rest of nex-agi/Nex-N2.5-Max has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 0 of 0 layers.
Tensors changed: Attention 27/27, Norms 12/12, MLP 43/43