MMFlash accelerates speculative decoding for vision-language models.
MMFlash targets visually dependent tokens in multimodal speculative decoding with sparse attention, yielding lossless speedups up to 3.58 times.
MMFlash targets visually dependent tokens in multimodal speculative decoding with sparse attention, yielding lossless speedups up to 3.58 times.
SparseEngine makes sparse attention core for LLM agents, covering 15 methods, cutting KV cache costs, speeding long-context inference and decoding.
MC-Sparse speeds diffusion transformers using training-free sparse attention, KV selection, query grouping, and residual caching for video and 3D.