usermma/DeepHat-V1-7B-Heretic-Abliterated
Partial fine-tune vs DeepHat/DeepHat-V1-7B. Partial change: 87% of the weights changed (attention, lm_head, mlp), in 26 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 78/84, Attention 104/196, lm_head 1/1
3 derivatives checked
- mradermacher/DeepHat-V1-7B-Heretic-Abliterated-GGUF: Same weights, converted
- mradermacher/DeepHat-V1-7B-Heretic-Abliterated-i1-GGUF: Quantization
- usermma/DeepHat-V1-7B-Heretic-Abliterated-mlx-8Bit: Quantization
1 other repos hold byte-for-byte identical weight files.