Linear-probe classification on SentEval
92.4MR Accuracycpt-text XL
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| cpt-text XLTraining Setting=Transfer from NLI, Model Size=XL (175B), Evaluation Protocol=Linear-probe2022.01 | 92.4 | 93.9 | 97 | 91.8 | 95.8 | 96.4 | 78.1 | 92.2 | |
| cpt-text XLTraining Setting=Unsupervised, Model Size=XL (175B), Evaluation Protocol=Linear-probe2022.01 | 92.2 | 93.5 | 97.4 | 91.5 | 96.2 | 97.4 | 74.1 | 91.8 | |
| cpt-text LTraining Setting=Transfer from NLI, Model Size=L, Evaluation Protocol=Linear-probe2022.01 | 90.8 | 93.5 | 96.2 | 91.2 | 95.7 | 96 | 76.9 | 91.5 | |
| cpt-text LTraining Setting=Unsupervised, Model Size=L, Evaluation Protocol=Linear-probe2022.01 | 90.6 | 92.6 | 97 | 90.6 | 95.3 | 97 | 73.6 | 90.9 | |
| cpt-text MTraining Setting=Transfer from NLI, Model Size=M, Evaluation Protocol=Linear-probe2022.01 | 89.8 | 92.7 | 95.7 | 91.3 | 95.3 | 96.6 | 76.5 | 91.1 | |
| cpt-text MTraining Setting=Unsupervised, Model Size=M, Evaluation Protocol=Linear-probe2022.01 | 89 | 90.9 | 96.7 | 89.6 | 93.9 | 96.6 | 73.6 | 89.9 | |
| SimCSETraining Setting=Transfer from NLI, Evaluation Protocol=Linear-probe2022.01 | 88.4 | 92.5 | 95.2 | 90.1 | 93.3 | 93.8 | 77.7 | 90.2 | |
| cpt-text STraining Setting=Transfer from NLI, Model Size=S (300M), Evaluation Protocol=Linear-probe2022.01 | 87.3 | 91 | 94.6 | 90.5 | 91.4 | 95 | 75.6 | 89.3 | |
| cpt-text STraining Setting=Unsupervised, Model Size=S (300M), Evaluation Protocol=Linear-probe2022.01 | 87.1 | 90.1 | 94.9 | 88.3 | 91.8 | 95.2 | 71.6 | 88.4 | |
| DECLUTRTraining Setting=Unsupervised, Evaluation Protocol=Linear-probe2022.01 | 85.2 | 90.7 | 95.8 | 88.5 | 90 | 93.2 | 74.6 | 88.3 | |
| SBERTTraining Setting=Transfer from NLI, Evaluation Protocol=Linear-probe2022.01 | 84.9 | 90.1 | 94.5 | 90.3 | 90.7 | 87.4 | 75.9 | 87.7 | |
| SimCSETraining Setting=Unsupervised, Evaluation Protocol=Linear-probe2022.01 | 84.7 | 88.6 | 95.4 | 87.5 | 89.5 | 95 | 72.4 | 87.6 | |
| BERTTraining Setting=Unsupervised, Evaluation Protocol=Linear-probe2022.01 | 78.7 | 86.2 | 94.4 | 88.7 | 84.4 | 92.8 | 69.4 | 84.9 |