NaniDAO/Llama-3.3-70B-Instruct-ablated
Abliterated vs meta-llama/Llama-3.3-70B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 80 of 80 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/161, MLP 80/240, Attention 80/320, lm_head 0/1
12 derivatives checked
- bartowski/Llama-3.3-70B-Instruct-ablated-GGUF: Quantization
- mradermacher/Llama-3.3-70B-Instruct-ablated-i1-GGUF: Quantization
- mradermacher/Llama-3.3-70B-Instruct-ablated-GGUF: Quantization
- Justbackup/Llama-3.3-70B-Instruct-ablated-GGUF: Quantization
- tensorblock/Llama-3.3-70B-Instruct-ablated-GGUF: Quantization
- DevQuasar-6/NaniDAO.Llama-3.3-70B-Instruct-ablated-GGUF: Quantization
- TareksGraveyard/Protobase-SCE1-LLaMa-70B: Full fine-tune
- SteelStorage/L3.1x3.3-MS-Hytroblated-70B-v2: Full fine-tune
- Kirkito/L3-R1-Framework-70B: Full fine-tune
- TareksGraveyard/Experimental-Base-V2.1-R1-LLaMa-70B: Full fine-tune
- TareksGraveyard/Protobase-SCE2-LLaMa-70B: Full fine-tune
- TareksGraveyard/Experimental-Base-V2-R1-LLaMa-70B: Full fine-tune
1 other repos hold byte-for-byte identical weight files.