# Open-set knowledge retrieval on T-REx (Hard)

Source: [Open-set knowledge retrieval on T-REx (Hard)](<https://www.sota2.com/research/sota/open-set-knowledge-retrieval-on-t-rex-hard>)

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

Data updated: 2026-02-27T08:17:49.951Z

Evaluation of model performance on the hard subset of the T-REx dataset for open-set knowledge retrieval, measured using Macro-averaged EM.

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 | Open-set knowledge retrieval |
| Dataset (including any stated split) | T-REx (Hard) |
| Metric | Macro EM |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 19 of 19 |
| 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

- [Macro EM](<https://www.sota2.com/research/sota/open-set-knowledge-retrieval-on-t-rex-hard.md?metric=Macro%20EM>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Macro EM | Result date | Report paper |
| --- | --- | --- | --- |
| NPM | 32.1 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + GPT-3 175B | 31.3 | Not recorded | 68fa6f7f1e209131cea45f9d |
| GPT-3 175B | 24.7 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + T5 | 22.4 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + GPT-3 13B | 22.4 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + T5 3B | 21.8 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + OPT 13B | 19.3 | Not recorded | 68fa6f7f1e209131cea45f9d |
| GPT-3 13B | 15.5 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + GPT-3 6.7B | 15.1 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + OPT 6.7B | 14.9 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + OPT 2.7B | 13.8 | Not recorded | 68fa6f7f1e209131cea45f9d |
| OPT 13B | 12.7 | Not recorded | 68fa6f7f1e209131cea45f9d |
| T5 3B | 11.5 | Not recorded | 68fa6f7f1e209131cea45f9d |
| T5 | 10.7 | Not recorded | 68fa6f7f1e209131cea45f9d |
| OPT 6.7B | 10.7 | Not recorded | 68fa6f7f1e209131cea45f9d |
| OPT 2.7B | 8.3 | Not recorded | 68fa6f7f1e209131cea45f9d |
| GPT-3 6.7B | 6.7 | Not recorded | 68fa6f7f1e209131cea45f9d |
| GPT-3 2.7B | 3.8 | Not recorded | 68fa6f7f1e209131cea45f9d |
| BM25 + GPT-3 2.7B | 3.6 | Not recorded | 68fa6f7f1e209131cea45f9d |

### NPM

Macro EM: 32.1. Result ID: 69a1531a35dfcabb00e645fb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Implementation](<https://www.sota2.com/github.com/facebookresearch/NPM>)

Recorded evaluation configuration: {"#Params":"1.0x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + GPT-3 175B

Macro EM: 31.3. Result ID: 69a1531a35dfcabb00e645fa.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa680c1e209131cea19d53.md>)

[Implementation](<https://github.com/openai/gpt-3/blob/master/overlap_frequency.md>)

Recorded evaluation configuration: {"#Params":"500x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### GPT-3 175B

Macro EM: 24.7. Result ID: 69a1531a35dfcabb00e645f3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a201e209131cea261f0.md>)

Recorded evaluation configuration: {"#Params":"500x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + T5

Macro EM: 22.4. Result ID: 69a1531a35dfcabb00e645eb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa709e1e209131cea4c470.md>)

[Implementation](<https://github.com/google-research/text-to-text-transfer-transformer>)

Recorded evaluation configuration: {"#Params":"2.2x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + GPT-3 13B

Macro EM: 22.4. Result ID: 69a1531a35dfcabb00e645f9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa680c1e209131cea19d53.md>)

[Implementation](<https://github.com/openai/gpt-3/blob/master/overlap_frequency.md>)

Recorded evaluation configuration: {"#Params":"37x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + T5 3B

Macro EM: 21.8. Result ID: 69a1531a35dfcabb00e645ec.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa709e1e209131cea4c470.md>)

[Implementation](<https://github.com/google-research/text-to-text-transfer-transformer>)

Recorded evaluation configuration: {"#Params":"8.5x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + OPT 13B

Macro EM: 19.3. Result ID: 69a1531a35dfcabb00e645f8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"#Params":"37x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### GPT-3 13B

Macro EM: 15.5. Result ID: 69a1531a35dfcabb00e645f2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e031e209131cea3d6a3.md>)

Recorded evaluation configuration: {"#Params":"37x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + GPT-3 6.7B

Macro EM: 15.1. Result ID: 69a1531a35dfcabb00e645f7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa680c1e209131cea19d53.md>)

[Implementation](<https://github.com/openai/gpt-3/blob/master/overlap_frequency.md>)

Recorded evaluation configuration: {"#Params":"19x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + OPT 6.7B

Macro EM: 14.9. Result ID: 69a1531a35dfcabb00e645f6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"#Params":"19x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + OPT 2.7B

Macro EM: 13.8. Result ID: 69a1531a35dfcabb00e645f4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"#Params":"7.6x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

### OPT 13B

Macro EM: 12.7. Result ID: 69a1531a35dfcabb00e645f1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"#Params":"37x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### T5 3B

Macro EM: 11.5. Result ID: 69a1531a35dfcabb00e645ea.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa709e1e209131cea4c470.md>)

[Implementation](<https://github.com/google-research/text-to-text-transfer-transformer>)

Recorded evaluation configuration: {"#Params":"8.5x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### T5

Macro EM: 10.7. Result ID: 69a1531a35dfcabb00e645e9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa709e1e209131cea4c470.md>)

[Implementation](<https://github.com/google-research/text-to-text-transfer-transformer>)

Recorded evaluation configuration: {"#Params":"2.2x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### OPT 6.7B

Macro EM: 10.7. Result ID: 69a1531a35dfcabb00e645ef.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"#Params":"19x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### OPT 2.7B

Macro EM: 8.3. Result ID: 69a1531a35dfcabb00e645ed.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"#Params":"7.6x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### GPT-3 6.7B

Macro EM: 6.7. Result ID: 69a1531a35dfcabb00e645f0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e031e209131cea3d6a3.md>)

Recorded evaluation configuration: {"#Params":"19x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### GPT-3 2.7B

Macro EM: 3.8. Result ID: 69a1531a35dfcabb00e645ee.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a201e209131cea261f0.md>)

Recorded evaluation configuration: {"#Params":"7.6x","C (Text Corpus)":"false","Zero-shot":"true"}. This record does not assert independent reproduction.

### BM25 + GPT-3 2.7B

Macro EM: 3.6. Result ID: 69a1531a35dfcabb00e645f5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6f7f1e209131cea45f9d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa680c1e209131cea19d53.md>)

[Implementation](<https://github.com/openai/gpt-3/blob/master/overlap_frequency.md>)

Recorded evaluation configuration: {"#Params":"7.6x","C (Text Corpus)":"true","Zero-shot":"true"}. This record does not assert independent reproduction.

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