# Trivia Question Answering on TriviaQA (comp. acc., bytes per sequence position)

Source: [Trivia Question Answering on TriviaQA (comp. acc., bytes per sequence position)](<https://www.sota2.com/research/sota/trivia-question-answering-on-triviaqa-comp-acc-bytes-per-sequence-position>)

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

Data updated: 2026-03-18T07:37:15.525Z

Evaluation of model performance on the TriviaQA dataset for Trivia Question Answering, reporting competitive accuracy and compression efficiency 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 | Trivia Question Answering |
| Dataset (including any stated split) | TriviaQA |
| Metric | Accuracy (Compression) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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

- [Accuracy (Compression)](<https://www.sota2.com/research/sota/trivia-question-answering-on-triviaqa-comp-acc-bytes-per-sequence-position.md?metric=Accuracy%20(Compression)>): Higher is better; unit: PERCENT.

- [Bytes per Sequence Position](<https://www.sota2.com/research/sota/trivia-question-answering-on-triviaqa-comp-acc-bytes-per-sequence-position.md?metric=Bytes%20per%20Sequence%20Position>): Lower is better; unit: SIZE.

## Recorded results

| Method | Accuracy (Compression) | Result date | Report paper |
| --- | --- | --- | --- |
| Llama | 69.5 | Not recorded | 69ba317b0636e5bf4afc071d |
| T-Free | 64.7 | Not recorded | 69ba317b0636e5bf4afc071d |
| HATified | 62.3 | Not recorded | 69ba317b0636e5bf4afc071d |

### Llama

Accuracy (Compression): 69.5. Result ID: 69ba561247f34ce924a3d986.

[Reporting paper](<https://www.sota2.com/research/paper/69ba317b0636e5bf4afc071d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75d11e209131cea6a028.md>)

[Implementation](<https://github.com/meta-llama/llama-agentic-system>)

Recorded evaluation configuration: {"shots":"5-shot"}. This record does not assert independent reproduction.

### T-Free

Accuracy (Compression): 64.7. Result ID: 69ba561247f34ce924a3d984.

[Reporting paper](<https://www.sota2.com/research/paper/69ba317b0636e5bf4afc071d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69ba317b0636e5bf4afc071d.md>)

Recorded evaluation configuration: {"shots":"5-shot"}. This record does not assert independent reproduction.

### HATified

Accuracy (Compression): 62.3. Result ID: 69ba561247f34ce924a3d985.

[Reporting paper](<https://www.sota2.com/research/paper/69ba317b0636e5bf4afc071d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69ba317b0636e5bf4afc071d.md>)

Recorded evaluation configuration: {"shots":"5-shot"}. This record does not assert independent reproduction.

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