kolibri1 TT: the routed-expert slot dequant read its fp8 source at twice the element stride - #3443

kolibri1 TT: the routed-expert slot dequant read its fp8 source at twice the element stride#3443

lu-zero merged 5 commits into

localai-org:mainlocalai-org/vllm.cpp:mainfrom

lu-zero:row/tt-kolibri-flips2lu-zero/vllm.cpp:row/tt-kolibri-flips2Copy head branch name to clipboard