experimentalmachines/QwenGrad-DPO
Full fine-tune (fewer layers) vs Qwen/Qwen3.5-2B. experimentalmachines/QwenGrad-DPO matches part of Qwen/Qwen3.5-2B (320 of its 632 tensors; the rest of Qwen/Qwen3.5-2B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 24 of 24 layers.
Tensors changed: Attention 132/168, Norms 39/79, Embeddings 1/1, MLP 72/72