# Natural Language Inference on NLI-bias

Source: [Natural Language Inference on NLI-bias](<https://www.sota2.com/research/sota/natural-language-inference-on-nli-bias>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T05:29:47.961Z

Evaluation of Natural Language Inference performance on the NLI-bias dataset, focusing on Accuracy and bias metrics.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Natural Language Inference |
| Dataset (including any stated split) | NLI-bias |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 10 of 10 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Accuracy](<https://www.sota2.com/research/sota/natural-language-inference-on-nli-bias.md?metric=Accuracy>): Higher is better; unit: PERCENT.

- [Bias](<https://www.sota2.com/research/sota/natural-language-inference-on-nli-bias.md?metric=Bias>): Lower is better; unit: SCALAR.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| MABEL | 86.4 | Not recorded | 68fa738e1e209131cea5d2e0 |
| Vanilla-tuning | 85.9 | Not recorded | 68fa738e1e209131cea5d2e0 |
| EAR | 85.9 | Not recorded | 68fa738e1e209131cea5d2e0 |
| MABEL | 81.3 | Not recorded | 68fa738e1e209131cea5d2e0 |
| EAR | 79.6 | Not recorded | 68fa738e1e209131cea5d2e0 |
| Vanilla-tuning | 79.5 | Not recorded | 68fa738e1e209131cea5d2e0 |
| Debiased-tuning | 77.4 | Not recorded | 68fa738e1e209131cea5d2e0 |
| Debiased-tuning | 75.1 | Not recorded | 68fa738e1e209131cea5d2e0 |
| ProSocialTuning | 74.7 | Not recorded | 68fa738e1e209131cea5d2e0 |
| ProSocialTuning | 73.8 | Not recorded | 68fa738e1e209131cea5d2e0 |

### MABEL

Accuracy: 86.4. Result ID: 6992ff533a6c31484912f098.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6dc51e209131cea3c093.md>)

Recorded evaluation configuration: {"Backbone":"RoBERTa-base"}. This record does not assert independent reproduction.

### Vanilla-tuning

Accuracy: 85.9. Result ID: 6992ff533a6c31484912f095.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

Recorded evaluation configuration: {"Backbone":"RoBERTa-base"}. This record does not assert independent reproduction.

### EAR

Accuracy: 85.9. Result ID: 6992ff533a6c31484912f097.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c101e209131cea31d71.md>)

Recorded evaluation configuration: {"Backbone":"RoBERTa-base"}. This record does not assert independent reproduction.

### MABEL

Accuracy: 81.3. Result ID: 6992ff533a6c31484912f093.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6dc51e209131cea3c093.md>)

Recorded evaluation configuration: {"Backbone":"BERT-base"}. This record does not assert independent reproduction.

### EAR

Accuracy: 79.6. Result ID: 6992ff533a6c31484912f092.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c101e209131cea31d71.md>)

Recorded evaluation configuration: {"Backbone":"BERT-base"}. This record does not assert independent reproduction.

### Vanilla-tuning

Accuracy: 79.5. Result ID: 6992ff533a6c31484912f090.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

Recorded evaluation configuration: {"Backbone":"BERT-base"}. This record does not assert independent reproduction.

### Debiased-tuning

Accuracy: 77.4. Result ID: 6992ff533a6c31484912f096.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

Recorded evaluation configuration: {"Backbone":"RoBERTa-base"}. This record does not assert independent reproduction.

### Debiased-tuning

Accuracy: 75.1. Result ID: 6992ff533a6c31484912f091.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

Recorded evaluation configuration: {"Backbone":"BERT-base"}. This record does not assert independent reproduction.

### ProSocialTuning

Accuracy: 74.7. Result ID: 6992ff533a6c31484912f094.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Implementation](<https://github.com/MSU-NLP-CSS/ProSocialTuning>)

Recorded evaluation configuration: {"Backbone":"BERT-base"}. This record does not assert independent reproduction.

### ProSocialTuning

Accuracy: 73.8. Result ID: 6992ff533a6c31484912f099.

[Reporting paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738e1e209131cea5d2e0.md>)

[Implementation](<https://github.com/MSU-NLP-CSS/ProSocialTuning>)

Recorded evaluation configuration: {"Backbone":"RoBERTa-base"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.