nisten/deepseek-0628-gguf

Same weights, converted vs deepseek-ai/DeepSeek-V2-Chat-0628. Same weights stored in another dtype (BF16 → BF16/F32), plus 177 new tensors.

Tensors changed: Embeddings 0/1, Norms 0/241, lm_head 0/1, Attention 0/300, MLP 0/239

On the Hub · Downloads on Model Pulse