AxionML/Kimi-K2.5-MXFP8

Partial fine-tune vs moonshotai/Kimi-K2.5. Partial change: 55% of the weights changed (attention, mlp), in 61 of 61 layers.

Tensors changed: Norms 0/357, MLP 789/1017, Attention 61/359, lm_head 0/1, Embeddings 0/4, other 0/58

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse