Repack of huihui-ai/Huihui-GLM-5.3-Flash-abliterated
Restored hot paths such as shared experts and output head of huihui-ai/Huihui-GLM-5.3-Flash-abliterated-GGUF/UD-Q2_K_XL and IQ2_XXS back to Q8, improving long form writing performance.
With <3GiB OS usage and MTP on, the Q2 version fits in a 128GB device with 0.5M context, and the IQ2 version fits with lower K cache precision for full 1M context.