Nickyang/Hy3-RAZOR-154B-A21B-E96of192
LoRA-style (fewer layers) vs tencent/Hy3. Nickyang/Hy3-RAZOR-154B-A21B-E96of192 matches part of tencent/Hy3 (23938 of its 47138 tensors; the rest of tencent/Hy3 has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, up_proj changed, in 80 of 81 layers; norms, biases and embeddings are untouched.
Tensors changed: MLP 997/1252, Attention 0/324, lm_head 0/1, other 0/1, Norms 0/328, Embeddings 0/1