Natural Language Inference on MNLI matched (val)
74.2AccuracyFloat32 Baseline
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Float32 BaselineModel=BERT-small, Protocol=Baseline2026.04 | 74.2 | — | |
| HCCSModel=BERT-small, Protocol=Retrained, Mode=INT16+DIV2026.04 | 72.3 | 1.9 | |
| Float32 BaselineModel=BERT-tiny, Protocol=Baseline2026.04 | 65.3 | — | |
| HCCSModel=BERT-tiny, Protocol=Retrained, Mode=INT16+DIV2026.04 | 63.9 | 1.3 | |
| HCCSModel=BERT-small, Protocol=No-retrain, Mode=INT16+DIV2026.04 | 60.2 | — | |
| HCCSModel=BERT-tiny, Protocol=No-retrain, Mode=INT16+DIV2026.04 | 48 | — |