Time-to-first-token (TTFT) on Llama-3.1-8B-Instruct 80K input length
9.2TTFT (s)TriangleMix + FlexPrefill
Evaluation Results
| Method | Links | |
|---|---|---|
| TriangleMix + FlexPrefill2025.07 | 9.2 | |
| TriangleMix + MInference2025.07 | 9.5 | |
| FlexPrefill2025.07 | 9.9 | |
| TriangleMix + XAttention2025.07 | 10 | |
| DuoAttention2025.07 | 11.7 | |
| TriangleMix2025.07 | 11.7 | |
| MInference2025.07 | 12.3 | |
| XAttention2025.07 | 12.4 | |
| Dense2025.07 | 15.9 |