SicariusSicariiStuff/Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct_Abliterated
Abliterated vs nvidia/Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 21 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/65, MLP 9/96, Attention 21/128, lm_head 0/1
9 derivatives checked
- SicariusSicariiStuff/Assistant_Pepe_8B: Full fine-tune
- mradermacher/Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct_Abliterated-i1-GGUF: Quantization
- mradermacher/Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct_Abliterated-GGUF: Same weights, converted
- OccultAI/Adversary-8B-v1.1: LoRA-style
- DarkArtsForge/Raven-8B-v1: LoRA-style
- EldritchLabs/Cthulhu-8B-v1.4: LoRA-style
- OccultAI/Morpheus-8B-v2: LoRA-style
- OccultAI/Morpheus-8B-v1: LoRA-style
- OccultAI/Barbot-8B-v1: LoRA-style
1 other repos hold byte-for-byte identical weight files.