# Commonsense Question Answering on CSQA Suite (Zero-Shot CS170K)

Source: [Commonsense Question Answering on CSQA Suite (Zero-Shot CS170K)](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k>)

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

Data updated: 2026-02-26T05:35:15.223Z

Evaluation of model performance across a suite of commonsense reasoning datasets (including ARC, BoolQ, HellaSwag, OBQA, PIQA, SciQ, WinoG) under a zero-shot setting, often referenced in the context of the CS170K evaluation framework.

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 | Commonsense Question Answering |
| Dataset (including any stated split) | CSQA (ARC-c, ARC-e, BoolQ, HellaSwag, OBQA, PIQA, SciQ, WinoG.) zero-shot CS170K |
| Metric | Average Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 4 of 4 |
| 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

- [Average Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=Average%20Accuracy>): Higher is better; unit: PERCENT.

- [ARC-Challenge Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=ARC-Challenge%20Accuracy>): Higher is better; unit: PERCENT.

- [ARC-Easy Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=ARC-Easy%20Accuracy>): Higher is better; unit: PERCENT.

- [BoolQ Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=BoolQ%20Accuracy>): Higher is better; unit: PERCENT.

- [HellaSwag Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=HellaSwag%20Accuracy>): Higher is better; unit: PERCENT.

- [OBQA Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=OBQA%20Accuracy>): Higher is better; unit: PERCENT.

- [PIQA Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=PIQA%20Accuracy>): Higher is better; unit: PERCENT.

- [SciQ Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=SciQ%20Accuracy>): Higher is better; unit: PERCENT.

- [WinoGrande Accuracy](<https://www.sota2.com/research/sota/commonsense-question-answering-on-csqa-suite-zero-shot-cs170k.md?metric=WinoGrande%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Average Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| QLoRA | 67.78 | Not recorded | 68fa78e91e209131cea7b2c5 |
| GSQ-Tuning | 67.73 | Not recorded | 68fa78e91e209131cea7b2c5 |
| GSQ-Tuning | 67.56 | Not recorded | 68fa78e91e209131cea7b2c5 |
| LLaMA2-7B | 64.13 | Not recorded | 68fa78e91e209131cea7b2c5 |

### QLoRA

Average Accuracy: 67.78. Result ID: 6995ae9a3badfaffe6adaea8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78e91e209131cea7b2c5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f6a1e209131cea456e1.md>)

[Implementation](<https://github.com/artidoro/qlora>)

Recorded evaluation configuration: {"LLMs branch bits":"4-16-16","low-rank branch bits":"16-16-16","Rank":64,"Zero-shot":"true"}. This record does not assert independent reproduction.

### GSQ-Tuning

Average Accuracy: 67.73. Result ID: 6995ae9a3badfaffe6adaea9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78e91e209131cea7b2c5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78e91e209131cea7b2c5.md>)

Recorded evaluation configuration: {"LLMs branch bits":"4-8-8","low-rank branch bits":"8-8-8","Rank":64,"Zero-shot":"true"}. This record does not assert independent reproduction.

### GSQ-Tuning

Average Accuracy: 67.56. Result ID: 6995ae9a3badfaffe6adaeaa.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78e91e209131cea7b2c5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78e91e209131cea7b2c5.md>)

Recorded evaluation configuration: {"LLMs branch bits":"4-6-6","low-rank branch bits":"6-6-6","Rank":64,"Zero-shot":"true"}. This record does not assert independent reproduction.

### LLaMA2-7B

Average Accuracy: 64.13. Result ID: 6995ae9a3badfaffe6adaea7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78e91e209131cea7b2c5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa701f1e209131cea492fe.md>)

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

Recorded evaluation configuration: {"LLMs branch bits":"16-16-None","low-rank branch bits":"w/o","Rank":64,"Zero-shot":"true"}. This record does not assert independent reproduction.

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