Question Answering on NarrativeQA (Prefill TP, Cache Hit, F1)
24,686.84Prefill Throughput (tok/s)ContextPilot w/o Annotations
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ContextPilot w/o AnnotationsHardware=32×H202025.11 | 24,686.84 | 35.71 | 40.38 | |
| ContextPilotHardware=32×H20, Annotations=enabled2025.11 | 24,686.84 | 35.71 | 41.08 | |
| VanillaHardware=32×H202025.11 | 16,247.32 | 5.14 | 40.2 | |
| ContextPilot w/o AnnotationsHardware=16×H202025.11 | 13,201.52 | 38.24 | 40.38 | |
| ContextPilotHardware=16×H20, Annotations=enabled2025.11 | 13,201.52 | 38.24 | 41.08 | |
| VanillaHardware=16×H202025.11 | 8,687.98 | 6.08 | 40.2 |