lucebox+16LLM speculative inference server for heterogeneous hardware & consumer GPUscudacuda-kernelsdflashheterogeneous-computing+15
AngelSlim+9Model compression toolkit engineered for enhanced usability, comprehensiveness, and efficiency.audiodeepseekdflashdiffusion+9
AngelSpec+6A unified, torch-native training framework for MTP and block-parallel speculative decoding.d-cutdflashdflyinference-acceleration+1