# Prompt Compression on L-Eval (test)

Source: [Prompt Compression on L-Eval (test)](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test>)

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

Data updated: 2026-03-23T11:14:19.989Z

Evaluation of model performance on the test split of the L-Eval benchmark suite, focusing on Prompt Compression capabilities across various QA tasks (Coursera QA, QuALITY, SFictionQA, etc.).

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 | Prompt Compression |
| Dataset (including any stated split) | L-Eval (test) |
| Metric | Coursera QA Accuracy |
| 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

- [Coursera QA Accuracy](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test.md?metric=Coursera%20QA%20Accuracy>): Higher is better; unit: PERCENT.

- [QuALITY Accuracy](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test.md?metric=QuALITY%20Accuracy>): Higher is better; unit: PERCENT.

- [SFictionQA Accuracy](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test.md?metric=SFictionQA%20Accuracy>): Higher is better; unit: PERCENT.

- [TPO Score](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test.md?metric=TPO%20Score>): Higher is better; unit: PERCENT.

- [LongFQA Accuracy](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test.md?metric=LongFQA%20Accuracy>): Higher is better; unit: PERCENT.

- [Legal Contract QA Accuracy](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test.md?metric=Legal%20Contract%20QA%20Accuracy>): Higher is better; unit: PERCENT.

- [Average Score (L-Eval)](<https://www.sota2.com/research/sota/prompt-compression-on-l-eval-test.md?metric=Average%20Score%20(L-Eval)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Coursera QA Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| LLMLingua-2 | 64.4 | Not recorded | 69c0c8d70636e5bf4afc394a |
| BEAVER | 64.4 | Not recorded | 69c0c8d70636e5bf4afc394a |
| Simply-BEAVER | 63.1 | Not recorded | 69c0c8d70636e5bf4afc394a |
| LongLLMLingua | 62.2 | Not recorded | 69c0c8d70636e5bf4afc394a |
| LLMLingua | 58.9 | Not recorded | 69c0c8d70636e5bf4afc394a |

### LLMLingua-2

Coursera QA Accuracy: 64.4. Result ID: 69c1208247f34ce924a97e4c.

[Reporting paper](<https://www.sota2.com/research/paper/69c0c8d70636e5bf4afc394a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74701e209131cea61cb2.md>)

[Implementation](<https://aka.ms/LLMLingua-2>)

Recorded evaluation configuration: {"Tokens":"2,110","1/τ":"4×","Compression Budget":"2,000"}. This record does not assert independent reproduction.

### BEAVER

Coursera QA Accuracy: 64.4. Result ID: 69c1208247f34ce924a97e4e.

[Reporting paper](<https://www.sota2.com/research/paper/69c0c8d70636e5bf4afc394a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c0c8d70636e5bf4afc394a.md>)

[Implementation](<https://github.com/JusperLee/BEAVER>)

Recorded evaluation configuration: {"Tokens":"2,180","1/τ":"4×","Compression Budget":"2,000"}. This record does not assert independent reproduction.

### Simply-BEAVER

Coursera QA Accuracy: 63.1. Result ID: 69c1208247f34ce924a97e4d.

[Reporting paper](<https://www.sota2.com/research/paper/69c0c8d70636e5bf4afc394a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c0c8d70636e5bf4afc394a.md>)

[Implementation](<https://github.com/JusperLee/BEAVER>)

Recorded evaluation configuration: {"Tokens":"2,123","1/τ":"4×","Compression Budget":"2,000"}. This record does not assert independent reproduction.

### LongLLMLingua

Coursera QA Accuracy: 62.2. Result ID: 69c1208247f34ce924a97e4b.

[Reporting paper](<https://www.sota2.com/research/paper/69c0c8d70636e5bf4afc394a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74701e209131cea61c6d.md>)

[Implementation](<https://aka.ms/LLMLingua>)

Recorded evaluation configuration: {"Tokens":"2,196","1/τ":"4×","Compression Budget":"2,000"}. This record does not assert independent reproduction.

### LLMLingua

Coursera QA Accuracy: 58.9. Result ID: 69c1208247f34ce924a97e4a.

[Reporting paper](<https://www.sota2.com/research/paper/69c0c8d70636e5bf4afc394a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71621e209131cea5103a.md>)

Recorded evaluation configuration: {"Tokens":"2,122","1/τ":"4×","Compression Budget":"2,000"}. This record does not assert independent reproduction.

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