apetersson/DeepSeek-V4-Flash-0731-Abliterated-FP8

LoRA-style vs deepseek-ai/DeepSeek-V4-Flash-0731. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of wo_b changed, in 33 of 43 layers; norms, biases and embeddings are untouched.

Tensors changed: Embeddings 0/1, MLP 0/1055, Attention 36/642, Norms 0/249, other 0/11

5 derivatives checked

On the Hub · Downloads on Model Pulse