# Question Answering on NQ-Open first 200 examples public (val)

Source: [Question Answering on NQ-Open first 200 examples public (val)](<https://www.sota2.com/research/sota/question-answering-on-nq-open-first-200-examples-public-val>)

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

Data updated: 2026-06-19T07:23:12.397Z

Evaluation of Question Answering performance specifically on the first 200 public examples of the NQ-Open validation set, measured using Mean Jaccard and Time-To-First-Token (TTFT) 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 | Question Answering |
| Dataset (including any stated split) | NQ-Open first 200 examples public (val) |
| Metric | Mean Jaccard |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 2 of 2 |
| 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

- [Mean Jaccard](<https://www.sota2.com/research/sota/question-answering-on-nq-open-first-200-examples-public-val.md?metric=Mean%20Jaccard>): Higher is better; unit: PERCENT.

- [Median TTFT (Retrieval Order)](<https://www.sota2.com/research/sota/question-answering-on-nq-open-first-200-examples-public-val.md?metric=Median%20TTFT%20(Retrieval%20Order)>): Lower is better; unit: TIME.

- [Median TTFT (Optimized Order)](<https://www.sota2.com/research/sota/question-answering-on-nq-open-first-200-examples-public-val.md?metric=Median%20TTFT%20(Optimized%20Order)>): Lower is better; unit: TIME.

## Recorded results

| Method | Mean Jaccard | Result date | Report paper |
| --- | --- | --- | --- |
| Retrieval Order | 6.8 | Not recorded | 6a34ccf2206542251d7c5e33 |
| CacheWeaver | 6.8 | Not recorded | 6a34ccf2206542251d7c5e33 |

### Retrieval Order

Mean Jaccard: 6.8. Result ID: 6a34edf5fbc373592c31d049.

[Reporting paper](<https://www.sota2.com/research/paper/6a34ccf2206542251d7c5e33.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69b91e209131cea23c26.md>)

[Implementation](<https://github.com/huggingface/transformers/blob/master/examples/rag/>)

Recorded evaluation configuration: {"Locality source":"TF-IDF QA-record reuse","Config":"C"}. This record does not assert independent reproduction.

### CacheWeaver

Mean Jaccard: 6.8. Result ID: 6a34edf5fbc373592c31d04a.

[Reporting paper](<https://www.sota2.com/research/paper/6a34ccf2206542251d7c5e33.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a34ccf2206542251d7c5e33.md>)

Recorded evaluation configuration: {"Locality source":"TF-IDF QA-record reuse","Config":"C"}. This record does not assert independent reproduction.

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