Question Answering on SQuAD v2
86.3AccuracyNoCache
Evaluation results
| Method | Links | ||
|---|---|---|---|
| NoCacheHardware=NVIDIA A100 GPU, Batch size=1, Sequence length=1282025.12 | 86.3 | — | |
| LLMCacheHardware=NVIDIA A100 GPU, Batch size=1, Sequence length=1282025.12 | 86.1 | — | |
| DocCacheHardware=NVIDIA A100 GPU, Batch size=1, Sequence length=1282025.12 | 85.8 | — | |
| SRMTraining=compute-matched, Training dataset=FineWeb-edu2026.05 | 16.87 | — | |
| TransformerTraining=compute-matched, Training dataset=FineWeb-edu2026.05 | 5.1 | — | |
| MambaTraining=compute-matched, Training dataset=FineWeb-edu2026.05 | 1.69 | — |