Attention Operator Performance on MLA 128 Head Dimension Sequence Length 1k
78.6TFLOPSLLM-TL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLM-TLHead Dimension=128, Mask=Causal, GPU=A100, Base Model=DeepSeek-V32025.06 | 78.6 | 4.25 | |
| cuDNNHead Dimension=128, Mask=Causal, GPU=A1002025.06 | 48.6 | — | |
| torchHead Dimension=128, Mask=Causal, GPU=A1002025.06 | 28.7 | — | |
| DeepSeek-V3Head Dimension=128, Mask=Causal, GPU=A1002025.06 | 18.5 | — |