# Commonsense on HellaSwag 10-shot

Source: [Commonsense on HellaSwag 10-shot](<https://www.sota2.com/research/sota/commonsense-on-hellaswag-10-shot>)

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

Data updated: 2026-03-04T12:24:40.254Z

Evaluation of commonsense reasoning capabilities on the HellaSwag dataset using a 10-shot prompting methodology, measured by Score.

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 |
| Dataset (including any stated split) | HellaSwag 10-shot |
| Metric | Accuracy (10-shot) |
| 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

- [Accuracy (10-shot)](<https://www.sota2.com/research/sota/commonsense-on-hellaswag-10-shot.md?metric=Accuracy%20(10-shot)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (10-shot) | Result date | Report paper |
| --- | --- | --- | --- |
| TuluDPO | 64.85 | Not recorded | 691aabe56750be83fe5ce08b |
| UltraMix-190k | 64.82 | Not recorded | 691aabe56750be83fe5ce08b |
| UltraMix-187k | 64.02 | Not recorded | 691aabe56750be83fe5ce08b |
| UltraMix-190k | 63.89 | Not recorded | 691aabe56750be83fe5ce08b |
| UltraMix-170k | 63.59 | Not recorded | 691aabe56750be83fe5ce08b |
| UltraMix-187k | 63.54 | Not recorded | 691aabe56750be83fe5ce08b |
| UltraMix-170k | 63.43 | Not recorded | 691aabe56750be83fe5ce08b |
| TuluDPO | 62.7 | Not recorded | 691aabe56750be83fe5ce08b |
| SFT | 60.85 | Not recorded | 691aabe56750be83fe5ce08b |
| SFT | 60.41 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (UM-190k) | 58.74 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (UM-187k) | 58.56 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (UM-170k) | 58.23 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (TuluDPO) | 57.51 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (ORPO) | 56.72 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (UltraFB) | 56.3 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (HelpSteer) | 56.24 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (SFT) | 56.2 | Not recorded | 691aabe56750be83fe5ce08b |
| Instella-3B-SFT (CodePref) | 55.99 | Not recorded | 691aabe56750be83fe5ce08b |

### TuluDPO

Accuracy (10-shot): 64.85. Result ID: 69a824736dbb7075b94e3f58.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78121e209131cea76a04.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-8B-TuluSFT","Training Mixture/Protocol":"TuluDPO"}. This record does not assert independent reproduction.

### UltraMix-190k

Accuracy (10-shot): 64.82. Result ID: 69a824736dbb7075b94e3f5b.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-8B-TuluSFT","Training Mixture/Protocol":"UM-190k"}. This record does not assert independent reproduction.

### UltraMix-187k

Accuracy (10-shot): 64.02. Result ID: 69a824736dbb7075b94e3f5a.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-8B-TuluSFT","Training Mixture/Protocol":"UM-187k"}. This record does not assert independent reproduction.

### UltraMix-190k

Accuracy (10-shot): 63.89. Result ID: 69a824736dbb7075b94e3f60.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"Base Model":"Qwen-2.5-7B-TuluSFT","Training Mixture/Protocol":"UM-190k"}. This record does not assert independent reproduction.

### UltraMix-170k

Accuracy (10-shot): 63.59. Result ID: 69a824736dbb7075b94e3f5e.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"Base Model":"Qwen-2.5-7B-TuluSFT","Training Mixture/Protocol":"UM-170k"}. This record does not assert independent reproduction.

### UltraMix-187k

Accuracy (10-shot): 63.54. Result ID: 69a824736dbb7075b94e3f5f.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"Base Model":"Qwen-2.5-7B-TuluSFT","Training Mixture/Protocol":"UM-187k"}. This record does not assert independent reproduction.

### UltraMix-170k

Accuracy (10-shot): 63.43. Result ID: 69a824736dbb7075b94e3f59.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-8B-TuluSFT","Training Mixture/Protocol":"UM-170k"}. This record does not assert independent reproduction.

### TuluDPO

Accuracy (10-shot): 62.7. Result ID: 69a824736dbb7075b94e3f5d.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78121e209131cea76a04.md>)

Recorded evaluation configuration: {"Base Model":"Qwen-2.5-7B-TuluSFT","Training Mixture/Protocol":"TuluDPO"}. This record does not assert independent reproduction.

### SFT

Accuracy (10-shot): 60.85. Result ID: 69a824736dbb7075b94e3f5c.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78121e209131cea76a04.md>)

Recorded evaluation configuration: {"Base Model":"Qwen-2.5-7B-TuluSFT","Training Mixture/Protocol":"SFT"}. This record does not assert independent reproduction.

### SFT

Accuracy (10-shot): 60.41. Result ID: 69a824736dbb7075b94e3f57.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78121e209131cea76a04.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-8B-TuluSFT","Training Mixture/Protocol":"SFT"}. This record does not assert independent reproduction.

### Instella-3B-SFT (UM-190k)

Accuracy (10-shot): 58.74. Result ID: 69a824736dbb7075b94e3e38.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"DPO Training Dataset":"UM-190k","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (UM-187k)

Accuracy (10-shot): 58.56. Result ID: 69a824736dbb7075b94e3e37.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"DPO Training Dataset":"UM-187k","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (UM-170k)

Accuracy (10-shot): 58.23. Result ID: 69a824736dbb7075b94e3e36.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

Recorded evaluation configuration: {"DPO Training Dataset":"UM-170k","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (TuluDPO)

Accuracy (10-shot): 57.51. Result ID: 69a824736dbb7075b94e3e31.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6916b7a30052cac1dc65e1ca.md>)

[Implementation](<https://github.com/AMD-AGI/Instella>)

Recorded evaluation configuration: {"DPO Training Dataset":"TuluDPO","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (ORPO)

Accuracy (10-shot): 56.72. Result ID: 69a824736dbb7075b94e3e32.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6916b7a30052cac1dc65e1ca.md>)

[Implementation](<https://github.com/AMD-AGI/Instella>)

Recorded evaluation configuration: {"DPO Training Dataset":"ORPO","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (UltraFB)

Accuracy (10-shot): 56.3. Result ID: 69a824736dbb7075b94e3e33.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6916b7a30052cac1dc65e1ca.md>)

[Implementation](<https://github.com/AMD-AGI/Instella>)

Recorded evaluation configuration: {"DPO Training Dataset":"UltraFB","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (HelpSteer)

Accuracy (10-shot): 56.24. Result ID: 69a824736dbb7075b94e3e34.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6916b7a30052cac1dc65e1ca.md>)

[Implementation](<https://github.com/AMD-AGI/Instella>)

Recorded evaluation configuration: {"DPO Training Dataset":"HelpSteer","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (SFT)

Accuracy (10-shot): 56.2. Result ID: 69a824736dbb7075b94e3e30.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6916b7a30052cac1dc65e1ca.md>)

[Implementation](<https://github.com/AMD-AGI/Instella>)

Recorded evaluation configuration: {"DPO Training Dataset":"None (SFT Baseline)","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

### Instella-3B-SFT (CodePref)

Accuracy (10-shot): 55.99. Result ID: 69a824736dbb7075b94e3e35.

[Reporting paper](<https://www.sota2.com/research/paper/691aabe56750be83fe5ce08b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6916b7a30052cac1dc65e1ca.md>)

[Implementation](<https://github.com/AMD-AGI/Instella>)

Recorded evaluation configuration: {"DPO Training Dataset":"CodePref","Evaluation Protocol":"10-shot"}. This record does not assert independent reproduction.

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