Common-sense QA on ARC Challenge
37.2AccuracyOPT-2.7B FP16
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OPT-2.7B FP16Model=OPT-2.7B, Precision=FP162023.08 | 37.2 | 8.96 | |
| TSLDModel=OPT-2.7B, Quantization=QAT-KD (tensor-wise)2023.08 | 33.45 | 11.05 | |
| GT+LogitModel=OPT-2.7B, Quantization=QAT-KD (tensor-wise)2023.08 | 32.16 | 12.04 | |
| LogitModel=OPT-2.7B, Quantization=QAT-KD (tensor-wise)2023.08 | 31.91 | 12.38 |