Weyaxi/Kimi-K3-NVFP4-W4A4

Partial fine-tune vs moonshotai/Kimi-K3. Partial change: 3% of the weights changed (lm_head, mlp), in 91 of 93 layers.

Tensors changed: Attention 0/1162, Norms 0/639, MLP 197/807, other 0/213, lm_head 1/1, Embeddings 0/3

On the Hub · Downloads on Model Pulse