general family · filed under inference & serving
Kernel fusion
Combining operations into fewer kernels to reduce intermediate work and keep hardware resources pipelined.
Also called inference kernel fusion, operator fusion.
- sources
- 2
- models
- 2
- labs adopt it
- 2
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1core 1
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
Through reasonable kernel fusion, we encapsulate the intricate operations and keep hardware resources fully pipelined inside a small number of fused kernels
coreunclearin DeepSeek-V4.1-FlashDeepSeek
optimization on areas such as operator fusion and communication optimization
usedunclearin Tencent Hy4 previewTencent
Filed alongside
Other methods under inference & serving :: inference kernel.
Batch-invariant deterministic kernelsCUDA GraphCUDA Graph capture size reductionFlashAttention 3MoE-side chunkingSynchronization-free static-shape MoE executionTensorRT-LLM multi-head attention backendAvoiding split-KDeepGEMM-based batch-invariant matrix multiplicationDistributed shared memory for cross-SM attention data exchangeDual-kernel batch-invariant attention decodingDynamic load balancingExp-free TopK kernelExpert-optimized Triton kernelsFA4 sheared-bias attention kernelFlashAttentionFlashAttention 4FlashKDAFP8 GEMMFused AttnRes merge and RMSNorm kernelFused latent down-projection and MoE-router GEMMFused mHC kernelsFused QSA kernelFused RoPE-attention-RoPE-cast kernel