nvidia/NV-Reason-CT
Full fine-tune (fewer layers) vs Qwen/Qwen3.5-4B. Qwen/Qwen3.5-4B matches part of nvidia/NV-Reason-CT (723 of its 1103 tensors; the rest of nvidia/NV-Reason-CT has no counterpart). On the shared part: Fine-tune: 93% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 1/4, MLP 96/196, Attention 216/320, Norms 104/203