mradermacher/sarashina2.2-3Bx4-moe-GGUF

Same weights, converted vs DataPilot/sarashina2.2-3Bx4-moe. Same weights stored in another dtype (BF16 → F16/F32), plus 96 new tensors.

Tensors changed: Embeddings 0/1, Norms 0/65, Attention 0/128, lm_head 0/1, other 0/32

On the Hub · Downloads on Model Pulse