mindlab-research/Macaron-V1-Preview-744B-Merged
LoRA-style vs zai-org/GLM-5.1. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, kv_a_proj_with_mqa, o_proj, q_a_proj, q_b_proj, up_proj changed, in 78 of 79 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/478, MLP 1707/1887, Attention 312/632, other 0/1