fix: avoid f16 overflow in Z-Image quantized matmuls on ROCm#1771

leejet merged 1 commit into

leejet:masterleejet/stable-diffusion.cpp:masterfrom

pwilkin:zimage-rocm-prec-f32pwilkin/stable-diffusion.cpp:zimage-rocm-prec-f32Copy head branch name to clipboard