minjaechoi/nemotron3-nano-30b-a3b-2p36bit-r16
LoRA-style vs nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, up_proj changed, in 23 of 52 layers; norms, biases and embeddings are untouched.
Tensors changed: other 0/207, Embeddings 0/1, Norms 0/76, MLP 2643/2691, Attention 0/24, lm_head 0/1