cantina-security/apex-flash-1-abliterated
Abliterated vs cantina-security/apex-flash-1. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 46 of 46 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 1/3, other 0/4, Norms 0/277, MLP 535/1888, Attention 46/827
6 derivatives checked
- freakyskittle/apex-flash-1-abliterated-GGUF: Quantization
- pqhaz/apex-flash-1-abliterated-GGUF: Quantization
- shamidk/apex-flash-1-abliterated-DS4-Q2-GGUF: Not comparable yet
- pqhaz/apex-flash-1-abliterated-FP8: Quantization
- Code4me2/apex-flash-1-abliterated-NVFP4: Quantization
- svbmrgd/apex-flash-1-abliterated-AWQ: Quantization