abacusai/Smaug-Flash
Partial fine-tune vs deepseek-ai/DeepSeek-V4-Flash-0731. Partial change: 24% of the weights changed (attention), in 43 of 43 layers.
Tensors changed: Embeddings 0/1, MLP 0/1845, Attention 257/893, Norms 0/249, other 0/12
Partial fine-tune vs deepseek-ai/DeepSeek-V4-Flash-0731. Partial change: 24% of the weights changed (attention), in 43 of 43 layers.
Tensors changed: Embeddings 0/1, MLP 0/1845, Attention 257/893, Norms 0/249, other 0/12