notSnix/Step-3.7-Flash-MTP-Draft-GGUF

Same weights, converted vs stepfun-ai/Step-3.7-Flash. Same weights stored in another dtype (BF16/F32 → BF16/F32), plus 16 new tensors.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/13, MLP 0/9, Attention 0/15

On the Hub · Downloads on Model Pulse