# Natural Language Understanding on GLUE (CoLA, MRPC, QNLI, MNLI, RTE Subset)

Source: [Natural Language Understanding on GLUE (CoLA, MRPC, QNLI, MNLI, RTE Subset)](<https://www.sota2.com/research/sota/natural-language-understanding-on-glue-cola-mrpc-qnli-mnli-rte-subset>)

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

Data updated: 2026-06-09T09:54:02.177Z

Evaluation of model performance across a specific subset of GLUE tasks: CoLA, MRPC, QNLI, MNLI, and RTE.

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 Understanding |
| Dataset (including any stated split) | GLUE |
| Metric | CoLA Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 12 of 12 |
| 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

- [CoLA Score](<https://www.sota2.com/research/sota/natural-language-understanding-on-glue-cola-mrpc-qnli-mnli-rte-subset.md?metric=CoLA%20Score>): Higher is better; unit: PERCENT.

- [MRPC Score](<https://www.sota2.com/research/sota/natural-language-understanding-on-glue-cola-mrpc-qnli-mnli-rte-subset.md?metric=MRPC%20Score>): Higher is better; unit: PERCENT.

- [QNLI Score](<https://www.sota2.com/research/sota/natural-language-understanding-on-glue-cola-mrpc-qnli-mnli-rte-subset.md?metric=QNLI%20Score>): Higher is better; unit: PERCENT.

- [MNLI Score](<https://www.sota2.com/research/sota/natural-language-understanding-on-glue-cola-mrpc-qnli-mnli-rte-subset.md?metric=MNLI%20Score>): Higher is better; unit: PERCENT.

- [RTE Score](<https://www.sota2.com/research/sota/natural-language-understanding-on-glue-cola-mrpc-qnli-mnli-rte-subset.md?metric=RTE%20Score>): Higher is better; unit: PERCENT.

- [Average GLUE Score](<https://www.sota2.com/research/sota/natural-language-understanding-on-glue-cola-mrpc-qnli-mnli-rte-subset.md?metric=Average%20GLUE%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | CoLA Score | Result date | Report paper |
| --- | --- | --- | --- |
| STAR | 66.62 | Not recorded | 6a279e7290966ade919bfb30 |
| No Interpolation (STAR ablation) | 66.02 | Not recorded | 6a279e7290966ade919bfb30 |
| Random basis (STAR ablation) | 65.8 | Not recorded | 6a279e7290966ade919bfb30 |
| STAR | 65.74 | Not recorded | 6a279e7290966ade919bfb30 |
| STAR | 65.54 | Not recorded | 6a279e7290966ade919bfb30 |
| DynMoE | 65.17 | Not recorded | 6a279e7290966ade919bfb30 |
| Cosine Router | 64.94 | Not recorded | 6a279e7290966ade919bfb30 |
| STAR | 64.69 | Not recorded | 6a279e7290966ade919bfb30 |
| No R (STAR ablation) | 64.57 | Not recorded | 6a279e7290966ade919bfb30 |
| Cosine Router | 64.12 | Not recorded | 6a279e7290966ade919bfb30 |
| Cosine Router | 64.1 | Not recorded | 6a279e7290966ade919bfb30 |
| Cosine Router | 63.63 | Not recorded | 6a279e7290966ade919bfb30 |

### STAR

CoLA Score: 66.62. Result ID: 6a27e0eefbc373592c251959.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

Recorded evaluation configuration: {"K":"8","k":"4","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### No Interpolation (STAR ablation)

CoLA Score: 66.02. Result ID: 6a27e0eefbc373592c25195f.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

Recorded evaluation configuration: {"K":"8","k":"4","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### Random basis (STAR ablation)

CoLA Score: 65.8. Result ID: 6a27e0eefbc373592c251960.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

Recorded evaluation configuration: {"K":"8","k":"4","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### STAR

CoLA Score: 65.74. Result ID: 6a27e0eefbc373592c25195d.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

Recorded evaluation configuration: {"K":"16","k":"4","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### STAR

CoLA Score: 65.54. Result ID: 6a27e0eefbc373592c251958.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

Recorded evaluation configuration: {"K":"8","k":"1","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### DynMoE

CoLA Score: 65.17. Result ID: 6a27e0eefbc373592c251955.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa775b1e209131cea72ab8.md>)

Recorded evaluation configuration: {"K":"9","k":"7.1","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### Cosine Router

CoLA Score: 64.94. Result ID: 6a27e0eefbc373592c251957.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa775b1e209131cea72ab8.md>)

Recorded evaluation configuration: {"K":"8","k":"4","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### STAR

CoLA Score: 64.69. Result ID: 6a27e0eefbc373592c25195c.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

Recorded evaluation configuration: {"K":"16","k":"1","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### No R (STAR ablation)

CoLA Score: 64.57. Result ID: 6a27e0eefbc373592c25195e.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

Recorded evaluation configuration: {"K":"8","k":"4","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### Cosine Router

CoLA Score: 64.12. Result ID: 6a27e0eefbc373592c25195b.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa775b1e209131cea72ab8.md>)

Recorded evaluation configuration: {"K":"16","k":"4","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### Cosine Router

CoLA Score: 64.1. Result ID: 6a27e0eefbc373592c251956.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa775b1e209131cea72ab8.md>)

Recorded evaluation configuration: {"K":"8","k":"1","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

### Cosine Router

CoLA Score: 63.63. Result ID: 6a27e0eefbc373592c25195a.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e7290966ade919bfb30.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa775b1e209131cea72ab8.md>)

Recorded evaluation configuration: {"K":"16","k":"1","Backbone":"BERT-large","Evaluation Protocol":"Fine-tuning"}. This record does not assert independent reproduction.

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