# Reasoning Generalization on Out-of-Distribution Avg

Source: [Reasoning Generalization on Out-of-Distribution Avg](<https://www.sota2.com/research/sota/reasoning-generalization-on-out-of-distribution-avg>)

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

Data updated: 2026-02-26T05:36:23.925Z

Evaluation of model performance on Reasoning Generalization using the average score derived from Out-of-Distribution samples.

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 | Reasoning Generalization |
| Dataset (including any stated split) | Out-of-Distribution Avg |
| Metric | Avg Score (OOD) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| 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

- [Avg Score (OOD)](<https://www.sota2.com/research/sota/reasoning-generalization-on-out-of-distribution-avg.md?metric=Avg%20Score%20(OOD)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Avg Score (OOD) | Result date | Report paper |
| --- | --- | --- | --- |
| TRAPO | 59.7 | Not recorded | 6940e7ac456146936284f900 |
| Fully Supervised | 57.3 | Not recorded | 6940e7ac456146936284f900 |
| Fully Supervised | 56.7 | Not recorded | 6940e7ac456146936284f900 |
| TRAPO | 56.1 | Not recorded | 6940e7ac456146936284f900 |
| Sentence-level Entropy | 52.6 | Not recorded | 6940e7ac456146936284f900 |
| TTRL | 52.4 | Not recorded | 6940e7ac456146936284f900 |
| Fully Supervised | 52.1 | Not recorded | 6940e7ac456146936284f900 |
| Sentence-level Entropy | 51.5 | Not recorded | 6940e7ac456146936284f900 |
| TTRL | 50.2 | Not recorded | 6940e7ac456146936284f900 |
| Token-level Entropy | 49.9 | Not recorded | 6940e7ac456146936284f900 |
| Token-level Entropy | 49.7 | Not recorded | 6940e7ac456146936284f900 |
| Self-certainty | 48.4 | Not recorded | 6940e7ac456146936284f900 |
| Self-certainty | 45.6 | Not recorded | 6940e7ac456146936284f900 |
| Qwen-Instruct | 43 | Not recorded | 6940e7ac456146936284f900 |
| Qwen-Base | 15.4 | Not recorded | 6940e7ac456146936284f900 |

### TRAPO

Avg Score (OOD): 59.7. Result ID: 699961bb5698665ec4ce1f8f.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Implementation](<https://github.com/ShenzhiYang2000/TRAPO>)

Recorded evaluation configuration: {"Training Paradigm":"Semi-supervised","Labeled Samples Count":"4K","Unlabeled Samples Count":"12K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Fully Supervised

Avg Score (OOD): 57.3. Result ID: 699961bb5698665ec4ce1f90.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

Recorded evaluation configuration: {"Training Paradigm":"Supervised","Labeled Samples Count":"45K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Fully Supervised

Avg Score (OOD): 56.7. Result ID: 699961bb5698665ec4ce1f8e.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

Recorded evaluation configuration: {"Training Paradigm":"Supervised","Labeled Samples Count":"4K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### TRAPO

Avg Score (OOD): 56.1. Result ID: 699961bb5698665ec4ce1f8d.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Implementation](<https://github.com/ShenzhiYang2000/TRAPO>)

Recorded evaluation configuration: {"Training Paradigm":"Semi-supervised","Labeled Samples Count":"1K","Unlabeled Samples Count":"3K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Sentence-level Entropy

Avg Score (OOD): 52.6. Result ID: 699961bb5698665ec4ce1f8c.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a61e209131cea79983.md>)

Recorded evaluation configuration: {"Training Paradigm":"Semi-supervised","Labeled Samples Count":"1K","Unlabeled Samples Count":"3K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### TTRL

Avg Score (OOD): 52.4. Result ID: 699961bb5698665ec4ce1f84.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76a91e209131cea6e3eb.md>)

Recorded evaluation configuration: {"Training Paradigm":"Unsupervised","Unlabeled Samples Count":"45K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Fully Supervised

Avg Score (OOD): 52.1. Result ID: 699961bb5698665ec4ce1f88.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

Recorded evaluation configuration: {"Training Paradigm":"Supervised","Labeled Samples Count":"1K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Sentence-level Entropy

Avg Score (OOD): 51.5. Result ID: 699961bb5698665ec4ce1f87.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a61e209131cea79983.md>)

Recorded evaluation configuration: {"Training Paradigm":"Unsupervised","Unlabeled Samples Count":"45K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### TTRL

Avg Score (OOD): 50.2. Result ID: 699961bb5698665ec4ce1f89.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76a91e209131cea6e3eb.md>)

Recorded evaluation configuration: {"Training Paradigm":"Semi-supervised","Labeled Samples Count":"1K","Unlabeled Samples Count":"3K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Token-level Entropy

Avg Score (OOD): 49.9. Result ID: 699961bb5698665ec4ce1f86.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a61e209131cea79983.md>)

Recorded evaluation configuration: {"Training Paradigm":"Unsupervised","Unlabeled Samples Count":"45K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Token-level Entropy

Avg Score (OOD): 49.7. Result ID: 699961bb5698665ec4ce1f8b.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a61e209131cea79983.md>)

Recorded evaluation configuration: {"Training Paradigm":"Semi-supervised","Labeled Samples Count":"1K","Unlabeled Samples Count":"3K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Self-certainty

Avg Score (OOD): 48.4. Result ID: 699961bb5698665ec4ce1f85.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a8d1e209131cea83dc9.md>)

Recorded evaluation configuration: {"Training Paradigm":"Unsupervised","Unlabeled Samples Count":"45K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Self-certainty

Avg Score (OOD): 45.6. Result ID: 699961bb5698665ec4ce1f8a.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a8d1e209131cea83dc9.md>)

Recorded evaluation configuration: {"Training Paradigm":"Semi-supervised","Labeled Samples Count":"1K","Unlabeled Samples Count":"3K","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Qwen-Instruct

Avg Score (OOD): 43. Result ID: 699961bb5698665ec4ce1f83.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74d81e209131cea6419f.md>)

[Implementation](<https://github.com/QwenLM/Qwen2-Math>)

Recorded evaluation configuration: {"Training Paradigm":"Original","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

### Qwen-Base

Avg Score (OOD): 15.4. Result ID: 699961bb5698665ec4ce1f82.

[Reporting paper](<https://www.sota2.com/research/paper/6940e7ac456146936284f900.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74d81e209131cea6419f.md>)

[Implementation](<https://github.com/QwenLM/Qwen2-Math>)

Recorded evaluation configuration: {"Training Paradigm":"Original","Backbone Model":"Qwen2.5-Math-7B","Sampling Temperature (T=0.6)":0.6}. This record does not assert independent reproduction.

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