🧰 FlashInfer

The items that name it 5

• ZipFusion: Fusing Lossless KV Decompression with Attention for Efficient LLM Decoding
  paper2026-10-24

  Attention library on top of which ZipFusion is implemented; baseline for the decode attention speedup comparison.

• [AINews] not much happened today
  article2026-10-03

  Kernel library receiving Prime Intellect's sparse-MLA kernel

• LFM2.5-VL-3B-DSpark (Liquid AI)
  article2026-09-24

  Attention backend used for the draft model in SGLang (--speculative-draft-attention-backend flashinfer).

• From CUDA to MLX: How K-Search Brings Decades of Kernel Expertise to Apple Silicon
  article2026-07-29

  Source of CUDA kernels (GQA decode, MLA decode, MLA prefill, MoE) used to evaluate the original K-Search search strategy

• vLLM v0.30.0
  repo

  Kernel library providing attention, MoE, all-reduce and KDA kernels (e.g. PCIe IPC all-reduce, XQA decode, CuTeDSL NVFP4)