ZichenAI/BaiHu-V1-Flash
Full fine-tune (fewer layers) vs Qwen/Qwen3-0.6B-Base. Qwen/Qwen3-0.6B-Base matches part of ZichenAI/BaiHu-V1-Flash (310 of its 506 tensors; the rest of ZichenAI/BaiHu-V1-Flash has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 1/1, Norms 71/113, MLP 84/84, Attention 112/112