random257/Qwen3.5-122B-A10B-trohrbaugh-Heretic-MTP

Same weights, converted vs trohrbaugh/Qwen3.5-122B-A10B-heretic. Same weights stored in another dtype (BF16 → BF16/F32), plus 200 new tensors.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/121, MLP 0/192, Attention 0/48

On the Hub · Downloads on Model Pulse