AnalysisAI ModelsJuly 15, 2026

Qwen 3.5 122B Heretic ROCmFP4 iMatrix

Community quantization achieves 28.45 tok/s on Strix Halo with 60.70 GiB memory. Created by RedParaglider as a ROCmFP4 iMatrix GGUF conversion.

1 source