amd/Kimi-K2-Thinking-MXFP4-AttnFP8

Partial fine-tune vs moonshotai/Kimi-K2-Thinking. Partial change: 69% of the weights changed (attention, mlp), in 61 of 61 layers.

Tensors changed: Norms 0/245, MLP 703/826, Attention 305/305, lm_head 0/1, Embeddings 0/1

On the Hub · Downloads on Model Pulse