magiccodingman/Qwen3.8-Flash-Next-Quark-MXFP4-fp8

Full fine-tune vs amd/Qwen3.8-Flash-Next-Quark-MXFP4. Fine-tune: 90% of the weights changed, across 2 of 48 layers.

Tensors changed: Norms 0/302, other 0/8, Attention 0/608, MLP 15/1925, Embeddings 128/137, lm_head 0/1

On the Hub · Downloads on Model Pulse