# Downstream Task Evaluation on LMEval Harness (MMLU, ARC-C, HellaSwag, TruthfulQA, Winogrande, GSM8K)

Source: [Downstream Task Evaluation on LMEval Harness (MMLU, ARC-C, HellaSwag, TruthfulQA, Winogrande, GSM8K)](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k>)

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

Data updated: 2026-02-26T05:50:01.160Z

Evaluation of model performance across a specific standard configuration of the LM Evaluation Harness, including MMLU, ARC Challenge, HellaSwag, TruthfulQA, Winogrande, and GSM8K.

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 | Downstream Task Evaluation |
| Dataset (including any stated split) | LM Evaluation Harness MMLU, ARC-Challenge, HellaSwag, TruthfulQA, Winogrande, GSM8K standard |
| Metric | MMLU |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 16 of 16 |
| 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

- [MMLU](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k.md?metric=MMLU>): Higher is better; unit: PERCENT.

- [ARC](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k.md?metric=ARC>): Higher is better; unit: PERCENT.

- [HellaSwag](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k.md?metric=HellaSwag>): Higher is better; unit: PERCENT.

- [TruthfulQA](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k.md?metric=TruthfulQA>): Higher is better; unit: PERCENT.

- [Winogrande](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k.md?metric=Winogrande>): Higher is better; unit: PERCENT.

- [GSM8K](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k.md?metric=GSM8K>): Higher is better; unit: PERCENT.

- [Average Score](<https://www.sota2.com/research/sota/downstream-task-evaluation-on-lmeval-harness-mmlu-arc-c-hellaswag-truthfulqa-winogrande-gsm8k.md?metric=Average%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | MMLU | Result date | Report paper |
| --- | --- | --- | --- |
| Initial Model | 65.8 | Not recorded | 698eb041c97a0b3e3b5e682c |
| DPO | 65.6 | Not recorded | 698eb041c97a0b3e3b5e682c |
| HyPO | 65 | Not recorded | 698eb041c97a0b3e3b5e682c |
| SimPO | 64.7 | Not recorded | 698eb041c97a0b3e3b5e682c |
| Initial Model | 63.7 | Not recorded | 698eb041c97a0b3e3b5e682c |
| HyPO | 63.6 | Not recorded | 698eb041c97a0b3e3b5e682c |
| DPO | 63.3 | Not recorded | 698eb041c97a0b3e3b5e682c |
| SimPO | 62.1 | Not recorded | 698eb041c97a0b3e3b5e682c |
| HyPO | 59.6 | Not recorded | 698eb041c97a0b3e3b5e682c |
| Initial Model | 59.2 | Not recorded | 698eb041c97a0b3e3b5e682c |
| Initial Model | 59 | Not recorded | 698eb041c97a0b3e3b5e682c |
| DPO | 58.8 | Not recorded | 698eb041c97a0b3e3b5e682c |
| HyPO | 58.5 | Not recorded | 698eb041c97a0b3e3b5e682c |
| SimPO | 58.4 | Not recorded | 698eb041c97a0b3e3b5e682c |
| SimPO | 58.1 | Not recorded | 698eb041c97a0b3e3b5e682c |
| DPO | 57.6 | Not recorded | 698eb041c97a0b3e3b5e682c |

### Initial Model

MMLU: 65.8. Result ID: 6998bc925698665ec41fad5f.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Instruct","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### DPO

MMLU: 65.6. Result ID: 6998bc925698665ec41fad60.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74511e209131cea6105d.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Instruct","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### HyPO

MMLU: 65. Result ID: 6998bc925698665ec41fad62.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Instruct","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### SimPO

MMLU: 64.7. Result ID: 6998bc925698665ec41fad61.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75901e209131cea68a5e.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Instruct","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### Initial Model

MMLU: 63.7. Result ID: 6998bc925698665ec41fad5b.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Base","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### HyPO

MMLU: 63.6. Result ID: 6998bc925698665ec41fad5e.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Base","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### DPO

MMLU: 63.3. Result ID: 6998bc925698665ec41fad5c.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74511e209131cea6105d.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Base","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### SimPO

MMLU: 62.1. Result ID: 6998bc925698665ec41fad5d.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75901e209131cea68a5e.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3","Model Type":"Base","Parameter Scale":"8B"}. This record does not assert independent reproduction.

### HyPO

MMLU: 59.6. Result ID: 6998bc925698665ec41fad5a.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Instruct","Parameter Scale":"7B"}. This record does not assert independent reproduction.

### Initial Model

MMLU: 59.2. Result ID: 6998bc925698665ec41fad57.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Instruct","Parameter Scale":"7B"}. This record does not assert independent reproduction.

### Initial Model

MMLU: 59. Result ID: 6998bc925698665ec41fad53.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Base","Parameter Scale":"7B"}. This record does not assert independent reproduction.

### DPO

MMLU: 58.8. Result ID: 6998bc925698665ec41fad58.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74511e209131cea6105d.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Instruct","Parameter Scale":"7B"}. This record does not assert independent reproduction.

### HyPO

MMLU: 58.5. Result ID: 6998bc925698665ec41fad56.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Base","Parameter Scale":"7B"}. This record does not assert independent reproduction.

### SimPO

MMLU: 58.4. Result ID: 6998bc925698665ec41fad55.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75901e209131cea68a5e.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Base","Parameter Scale":"7B"}. This record does not assert independent reproduction.

### SimPO

MMLU: 58.1. Result ID: 6998bc925698665ec41fad59.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75901e209131cea68a5e.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Instruct","Parameter Scale":"7B"}. This record does not assert independent reproduction.

### DPO

MMLU: 57.6. Result ID: 6998bc925698665ec41fad54.

[Reporting paper](<https://www.sota2.com/research/paper/698eb041c97a0b3e3b5e682c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74511e209131cea6105d.md>)

Recorded evaluation configuration: {"Backbone":"Mistral","Model Type":"Base","Parameter Scale":"7B"}. This record does not assert independent reproduction.

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