nvidia/Eagle2-2B
Full fine-tune (fewer layers) vs Qwen/Qwen2.5-1.5B-Instruct. Qwen/Qwen2.5-1.5B-Instruct matches part of nvidia/Eagle2-2B (337 of its 793 tensors; the rest of nvidia/Eagle2-2B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Norms 57/57, MLP 84/84, Attention 196/196
2 other repos hold byte-for-byte identical weight files.