amd/Kimi-K2.5-MXFP4

Partial fine-tune vs moonshotai/Kimi-K2.5. Partial change: 3% of the weights changed (mlp), in 60 of 61 layers.

Tensors changed: Norms 0/357, Attention 0/359, MLP 606/834, lm_head 0/1, Embeddings 0/4, other 0/58

On the Hub · Downloads on Model Pulse