PrimeIntellect/INTELLECT-3
Full fine-tune vs zai-org/GLM-4.5-Air-Base. Fine-tune: 100% of the weights changed, across 46 of 46 layers.
Tensors changed: Embeddings 1/1, Norms 93/93, MLP 2460/2505, Attention 322/322, lm_head 1/1
3 derivatives checked
- bartowski/PrimeIntellect_INTELLECT-3-GGUF: Quantization
- Ex0bit/Elbaz-Prime-Intellect-3_Prism_Abliterated: Quantization
- jtl11/INTELLECT-3-REAP-50-heretic: LoRA-style (fewer layers)
3 other repos hold byte-for-byte identical weight files.