Global Sliding Window Attention on Llama2-7b q=32, k=32 (1k)
67.36TFLOPSCuBridge
Evaluation Results
| Method | Links | |
|---|---|---|
| CuBridgeImplementation=CuBridge2026.05 | 67.36 | |
| FlexAttentionImplementation=PyTorch FlexAttention2026.05 | 56.96 | |
| Qimeng AttnImplementation=Qimeng Attention2026.05 | 29.82 | |
| TorchImplementation=Standard PyTorch2026.05 | 10.61 |