Sigmoid Attention on Llama2-7b (q=32, k=32) (1k)
134.56Throughput (TFLOPS)CuBridge
Evaluation Results
| Method | Links | |
|---|---|---|
| CuBridgeImplementation=CuBridge2026.05 | 134.56 | |
| Qimeng AttnImplementation=Qimeng Attention2026.05 | 45.05 | |
| TorchImplementation=Standard PyTorch2026.05 | 30.48 |