s-zaizen/DeepSeek-V4.1-Flash-Abliterated
LoRA-style vs deepseek-ai/DeepSeek-V4.1-Flash. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of wo_b changed, in 39 of 40 layers; norms, biases and embeddings are untouched.
Tensors changed: other 0/17, Attention 39/542, MLP 0/1149, Norms 0/248, Embeddings 0/6
1 derivatives checked
- audreyt/DeepSeek-V4.1-Flash-Abliterated-GGUF: Not comparable yet
2 other repos hold byte-for-byte identical weight files.