Question Answering on MMLU (4-way Zero-shot Transfer)
91.184AccuracyReinforcement Inference
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Reinforcement InferenceK (choices)=4, Entropy threshold τH (nats)=0.78, MSP threshold τMSP=0.4, Zero-shot transfer=true, Dataset size=140422026.02 | 91.184 | 16 | 11 | |
| BaselineK (choices)=4, Entropy threshold τH (nats)=0.78, MSP threshold τMSP=0.4, Zero-shot transfer=true, Dataset size=140422026.02 | 91.134 | — | — |