Attention Operator Throughput on Qwen2.5 72B (64 Q-heads/8 KV-heads/128 Head-dimension)
222.5Attention Throughput (TFLOPS)
flash-attn v2
Updated 4d ago
Evaluation Results
| Method | Links | |
|---|---|---|
2025.06 | 222.5 | |
2025.06 | 216 | |
2025.06 | 207.9 | |
2025.06 | 205.1 | |
2025.06 | 202.1 | |
2025.06 | 201.8 | |
2025.06 | 201.8 | |
2025.06 | 192.9 | |
2025.06 | 191.8 | |
2025.06 | 186.4 | |
2025.06 | 181.5 | |
2025.06 | 176.1 | |
2025.06 | 172.4 | |
2025.06 | 168.1 | |
2025.06 | 165.1 | |
2025.06 | 157.8 | |
2025.06 | 154.7 | |
2025.06 | 148.3 | |
2025.06 | 140.2 | |
2025.06 | 136.7 | |
2025.06 | 127.1 | |
2025.06 | 117.4 | |
2025.06 | 116.3 | |
2025.06 | 86.1 | |
2025.06 | 11.2 | |
2025.06 | 11.1 | |
2025.06 | 10.9 | |
2025.06 | 10.2 | |
2025.06 | 8.8 |