cerebras/GLM-4.6-REAP-268B-A32B
LoRA-style vs zai-org/GLM-4.6. 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 89 of 92 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/369, MLP 1098/1398, Attention 0/644, lm_head 0/1
2 derivatives checked
- unsloth/GLM-4.6-REAP-268B-A32B-GGUF: Same weights, converted
- ArliAI/GLM-4.6-REAP-268B-A32B-Derestricted: LoRA-style