opencl: transpose q4_K noshuffle scales to allow better coalesced reads for Adreno GPUs. - #25805

opencl: transpose q4_K noshuffle scales to allow better coalesced reads for Adreno GPUs.#25805

max-krasnyansky merged 1 commit into

ggml-org:masterggml-org/llama.cpp:masterfrom

qualcomm:hq/q4k-scales-transpose-prqualcomm/llama.cpp:hq/q4k-scales-transpose-prCopy head branch name to clipboard