Shifusen/Qwen3-Next-80B-A3B-Instruct-Decensored
LoRA-style vs Qwen/Qwen3-Next-80B-A3B-Instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate, gate_proj, in_proj_ba, in_proj_qkvz, k_proj, o_proj, out_proj, q_proj, shared_expert_gate, up_proj, v_proj changed, in 48 of 48 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/157, Attention 156/264, MLP 239/2514, lm_head 0/1