Jiunsong/SuperDeepseek-V4-Flash-abliterated

Partial fine-tune vs deepseek-ai/DeepSeek-V4-Flash-0731. Partial change: 21% of the weights changed (attention, other), in 43 of 43 layers.

Tensors changed: Embeddings 0/1, MLP 0/1055, Attention 46/642, Norms 0/249, other 1/11

1 derivatives checked