KernelSeek beats FlashAttention-3 by asking questions, not adding tools
A tool-rich agent stalled at 0.50x after 133 steps; KernelSeek reached 1.13x, then 1.16x and 1.46x on later kernels.
48
Product
FlashAttention-4
- Mentions · 7d0
- Mentions · 30d1
- All mentions3
- First seenSep 13, 2026
- Last mentioned7d ago
Timeline
3 articles · newest firstApproximating Softmax in Pretrained LLMs: Model Sensitivity and Kernel Acceleration
Approximating softmax in ten frozen LLMs yields a FlashAttention-4 kernel up to 25.8% faster on NVIDIA B200.
55
VC-Attention: Value Smoothing and Softmax Casting for Low-bit Attention
VC-Attention is a training-free low-bit attention method for diffusion transformers, achieving 1.46-1.59x kernel speedups on datacenter GPUs with higher fidelity.
38
Appears with
- NVIDIAvendor · 3
- HunyuanVideo-1.5ai model · 1
- Blackwell B200product · 1
- Wan2.2ai model · 1
- KernelSeekai model · 1
- LongCat-Videoai model · 1
- CUDAproduct · 1
- MiniMax-H3ai model · 1
- VC-Attentionproduct · 1
- FlashAttention-3product · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.