# Few-shot Natural Language Processing on MetaICL (test)

Source: [Few-shot Natural Language Processing on MetaICL (test)](<https://www.sota2.com/research/sota/few-shot-natural-language-processing-on-metaicl-test>)

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

Data updated: 2026-02-26T05:39:14.027Z

Evaluation of Few-shot Natural Language Processing performance on the test split of the MetaICL benchmark, reporting scores across various NLP subtasks (e.g., QA, NLI, Classification).

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 | Few-shot Natural Language Processing |
| Dataset (including any stated split) | MetaICL (test) |
| Metric | Classification Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Classification Score](<https://www.sota2.com/research/sota/few-shot-natural-language-processing-on-metaicl-test.md?metric=Classification%20Score>): Higher is better; unit: PERCENT.

- [LR Score](<https://www.sota2.com/research/sota/few-shot-natural-language-processing-on-metaicl-test.md?metric=LR%20Score>): Higher is better; unit: PERCENT.

- [QA Score](<https://www.sota2.com/research/sota/few-shot-natural-language-processing-on-metaicl-test.md?metric=QA%20Score>): Higher is better; unit: PERCENT.

- [NLI Score](<https://www.sota2.com/research/sota/few-shot-natural-language-processing-on-metaicl-test.md?metric=NLI%20Score>): Higher is better; unit: PERCENT.

- [Paraphrasing Score](<https://www.sota2.com/research/sota/few-shot-natural-language-processing-on-metaicl-test.md?metric=Paraphrasing%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Classification Score | Result date | Report paper |
| --- | --- | --- | --- |
| UnpredicTable-5k | 46.1 | Not recorded | 68fa6d351e209131cea3872e |
| MetaICL k-shot | 43.4 | Not recorded | 68fa6d351e209131cea3872e |
| MetaICL k-shot | 38.2 | Not recorded | 68fa6d351e209131cea3872e |
| GPT2 | 37.4 | Not recorded | 68fa6d351e209131cea3872e |
| GPT2 | 34.2 | Not recorded | 68fa6d351e209131cea3872e |

### UnpredicTable-5k

Classification Score: 46.1. Result ID: 698f842b3a6c314849f99464.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea3872e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea3872e.md>)

[Implementation](<https://github.com/JunShern/few-shot-adaptation>)

Recorded evaluation configuration: {"training\_data":"UnpredicTable-5k","backbone":"GPT2-large","shot\_setting":"k-shot"}. This record does not assert independent reproduction.

### MetaICL k-shot

Classification Score: 43.4. Result ID: 698f842b3a6c314849f99463.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea3872e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c7c1e209131cea34419.md>)

Recorded evaluation configuration: {"training\_data":"NLP (IID)"}. This record does not assert independent reproduction.

### MetaICL k-shot

Classification Score: 38.2. Result ID: 698f842b3a6c314849f99462.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea3872e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c7c1e209131cea34419.md>)

Recorded evaluation configuration: {"training\_data":"NLP (OOD)"}. This record does not assert independent reproduction.

### GPT2

Classification Score: 37.4. Result ID: 698f842b3a6c314849f99461.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea3872e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c7c1e209131cea34419.md>)

Recorded evaluation configuration: {"shot\_setting":"k-shot"}. This record does not assert independent reproduction.

### GPT2

Classification Score: 34.2. Result ID: 698f842b3a6c314849f99460.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6d351e209131cea3872e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c7c1e209131cea34419.md>)

Recorded evaluation configuration: {"shot\_setting":"0-shot"}. This record does not assert independent reproduction.

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