PKU-SEC-Lab/HybriMoE - official implementation of HybriMoE

GitHub repo titled [DAC 25] Official implement of HybriMoE. README read 2026-10-18. It is a fork of ktransformers (kvcache-ai/ktransformers as a submodule), Apache-2.0. The example command runs DeepSeek-V2-Lite-Chat from a Q4_K_M GGUF via ktransformers/local_chat.py with --cache_size 16 and --prefetch_size 0. The citation block in the README still points at the arXiv preprint, not the DAC DOI. Six commits on main at the time of access. This is a research fork, not an upstream ktransformers feature. Used only to confirm the code exists and runs on top of ktransformers, not as a source of the speedup numbers.