IST-DASLab/qmoe - CUDA kernels and compression code for QMoE

Apache-2.0. The README names the layout: sub1.py is the compressed-inference path, sub1_cuda_kernel.cu the CUDA kernels, switch.py the compression framework, test.py the per-layer benchmarks. Pinned dependencies are torch 2.0.0+cu117, transformers 4.28.0, datasets 2.10.1, CUDA 11.4 drivers. Compressed checkpoints are linked as ISTA-DASLab/switch-base-128_qmoe, switch-large-128_qmoe and switch-c-2048_qmoe. The page showed 4 commits on main, 279 stars and 25 forks on 2026-10-18; the commit dates themselves were not rendered, so no maintenance claim is made from the count.