No MTP yet, will update once supported upstream
Quant optimized for quality / speed on a Strix Halo 128GiB system. Possibly also beneficial on DGX Spark and similar systems. Refer to tensor types for the recipe.
This quant is deliberately larger than total resident memory as the ngram is expected to be mmapped to internal NVME. If you want a fully resident solution, I recommend instead just using https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF