MiniMax AI · AI Infrastructure
MiniMax Sparse Attention (MSA)
CUDA kernel library for NVIDIA SM100 GPUs shipping dense FlashAttention plus sparse top-k block attention via two JIT stacks (csrc + CuTe-DSL), with paged FP8 decode and BF16/FP8/NVFP4/FP4 quantization. A cheap proxy pass selects high-scoring attention blocks, then the sparse kernel computes only those — the attention powering MiniMax M3's 1M-context speedups.
Open the source ↗Similar Developer Tools releases
- DeepInfer
DeepSeek · Inference Engine · Week 44, 2026
- Toward Recursive Self-Improvement
Z.ai · Self-Built Inference Stack · Week 39, 2026
- A Hardware-Level Tour of LLM Inference
Dr Ashish Bamania · LLM Inference Internals · Week 36, 2026
- Kie.ai
Kie.ai · AI Infrastructure · Week 31, 2026
- WaveSpeedAI
WaveSpeedAI · Inference Engine · Week 31, 2026
- Exo
exo-explore · Inference Engine · Week 31, 2026
More from MiniMax AI
- MiniMax Code
MiniMax AI · Terminal Coding Agent · Week 39, 2026
- MiniMax Music 3
MiniMax AI · Music Generation · Week 33, 2026
- MiniMax-H3
MiniMax AI · Text-to-Video · Week 32, 2026
- MiniMax M3
MiniMax AI · Text Generation · Week 24, 2026
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.