elinas/Qwen2-11.3B
Partial fine-tune (fewer layers) vs Qwen/Qwen2-7B. Qwen/Qwen2-7B matches part of elinas/Qwen2-11.3B (339 of its 531 tensors; the rest of elinas/Qwen2-11.3B has no counterpart). On the shared part: Partial change: 18% of the weights changed (attention, mlp, norms), in 6 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 12/57, MLP 18/84, Attention 42/196, lm_head 0/1