ggml-hrx: MUL_MAT_ID at input sizes that are a multiple of 32 (gpt-oss experts on HRX) + decode input row stride - #66

ggml-hrx: MUL_MAT_ID at input sizes that are a multiple of 32 (gpt-oss experts on HRX) + decode input row stride#66

bong-water-water-bong merged 4 commits into

1bit/hrx-vulkan-patched1bit-MONSTER/llama.cpp:1bit/hrx-vulkan-patchedfrom

1bit/hrx-mmid-g32-v21bit-MONSTER/llama.cpp:1bit/hrx-mmid-g32-v2Copy head branch name to clipboard