dealignai/GLM-5.3-Flash-UNCENSORED-FP8
Abliterated vs zai-org/GLM-5.3-Flash. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 34 of 46 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 52/1840, lm_head 0/1, Embeddings 0/3, other 0/4, Norms 0/277, Attention 43/875
7 derivatives checked
- AliceThirty/GLM-5.3-Flash-UNCENSORED-GGUF: Quantization
- Solstice-AI/GLM-5.3-Flash-UNCENSORED-mlx-oQ4e-DFlash2: Different architecture
- Solstice-AI/GLM-5.3-Flash-UNCENSORED-mlx-oQ8e-DFlash2: Different architecture
- Solstice-AI/GLM-5.3-Flash-UNCENSORED-NVFP4-DFlash2: Different architecture
- Cypherfox/GLM-5.3-Flash-dealignai-orcarouter-merge: Quantization
- Solstice-AI/GLM-5.3-Flash-UNCENSORED-GGUF: Quantization
- junmo-k/GLM-5.3-Flash-UNCENSORED-JK-oQ3.5e: Different architecture
6 other repos hold byte-for-byte identical weight files.