mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated
Abliterated vs meta-llama/Llama-3.1-8B-Instruct. Looks like abliteration (refusal removal) done on a re-parametrized copy of the model: the output projections (down_proj, embed_tokens, o_proj) changed, in 32 of 32 layers, and the norm weights were re-scaled (x1.03 median) with the matrices after them (q/k/v, gate/up) shrinking by about the inverse, which changes no behavior by itself.
Tensors changed: Embeddings 1/1, Norms 0/65, MLP 32/96, Attention 32/128, lm_head 0/1
31 derivatives checked
- bartowski/Meta-Llama-3.1-8B-Instruct-abliterated-GGUF: Quantization
- legraphista/Meta-Llama-3.1-8B-Instruct-abliterated-IMat-GGUF: Same weights, converted
- mradermacher/Meta-Llama-3.1-8B-Instruct-abliterated-GGUF: Same weights, converted
- mradermacher/Meta-Llama-3.1-8B-Instruct-abliterated-i1-GGUF: Quantization
- QuantFactory/Thespis-Llama-3.1-8B-GGUF: Full fine-tune
- QuantFactory/llama3.1-cc-8B-GGUF: Quantization
- QuantFactory/Llama3.1-Allades-8B-GGUF: Quantization
- featherless-ai-quants/mlabonne-Meta-Llama-3.1-8B-Instruct-abliterated-GGUF: Quantization
- hakks660/Meta-Llama-3.1-8B-Instruct-abliterated-GGUF: Same weights, converted
- sinhal/SageV2-8B-LongContext-NSFW-RP: Full fine-tune
- solidrust/Meta-Llama-3.1-8B-Instruct-abliterated-AWQ: Quantization
- tensorblock/Meta-Llama-3.1-8B-Instruct-abliterated-GGUF: Quantization
- YorkieOH10/Meta-Llama-3.1-8B-Instruct-abliterated-Q8_0-GGUF: Quantization
- 0xBreath/Meta-Llama-3.1-8B-Instruct-abliterated-q8-mlx: Quantization
- Locutusque/Thespis-Llama-3.1-8B: Full fine-tune
- nbeerbower/llama3.1-cc-8B: LoRA-style
- netcat420/MFANN-llama3.1-abliterated-v2: LoRA-style
- netcat420/MFANN-SFT: LoRA-style
- netcat420/MFANN-Llama3.1-Abliterated-Slerp-TIES: LoRA-style
- netcat420/MFANN-Llama3.1-Abliterated-SLERP-V5: Partial fine-tune
- netcat420/MFANN-Llama3.1-Abliterated-SLERP-TIES-V3: LoRA-style
- netcat420/MFANN-llama3.1-Abliterated-SLERP: Partial fine-tune
- netcat420/MFANN-Llama3.1-Abliterated-Slerp-V3.2: Partial fine-tune
- netcat420/MFANN-llama3.1-abliterated-SLERP-v3: LoRA-style
- netcat420/MFANNv0.20.12: LoRA-style
- netcat420/MFANNv0.21.10: Partial fine-tune
- yuvraj17/Llama3-8B-abliterated-Spectrum-slerp: Partial fine-tune
- netcat420/MFANNv0.20.10: Partial fine-tune
- brianmatzelle/llama3.1-8b-instruct-hasanpiker-abliterated: LoRA-style
- nicoboss/Meta-Llama-3.1-8B-Instruct-abliterated-Sabresooth: LoRA-style
2 other repos hold byte-for-byte identical weight files.