ewin-reg/WeMM-Embedding-2B-INT8

Same weights, converted vs ewin-reg/WeMM-Embedding-2B-Quantized. Same weights stored in another dtype (F8_E4M3/U8 → I8/U8), plus 139 new tensors.

Tensors changed: Norms 0/177, Attention 0/312, Embeddings 0/4, MLP 0/270

On the Hub · Downloads on Model Pulse