# Synthetic in-context reasoning on MAD

Source: [Synthetic in-context reasoning on MAD](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad>)

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

Data updated: 2026-05-20T07:38:24.236Z

Evaluation of model performance on the Synthetic in-context reasoning task using the MAD dataset, measured by metrics like Compress, Fuzzy Recall, and In-Ctx Recall.

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 | Synthetic in-context reasoning |
| Dataset (including any stated split) | MAD |
| Metric | Compress |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Compress](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad.md?metric=Compress>): Higher is better; unit: PERCENT.

- [Fuzzy Recall](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad.md?metric=Fuzzy%20Recall>): Higher is better; unit: PERCENT.

- [In-Ctx Recall](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad.md?metric=In-Ctx%20Recall>): Higher is better; unit: PERCENT.

- [Memorize TrainSet](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad.md?metric=Memorize%20TrainSet>): Higher is better; unit: PERCENT.

- [Noisy Recall](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad.md?metric=Noisy%20Recall>): Higher is better; unit: PERCENT.

- [Selective Copy](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad.md?metric=Selective%20Copy>): Higher is better; unit: PERCENT.

- [Average Score](<https://www.sota2.com/research/sota/synthetic-in-context-reasoning-on-mad.md?metric=Average%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Compress | Result date | Report paper |
| --- | --- | --- | --- |
| FLASH PD-SSM | 58 | Not recorded | 6a0d3ff5f984739ed67376b4 |
| Mamba | 52.7 | Not recorded | 6a0d3ff5f984739ed67376b4 |
| Transformer | 51.6 | Not recorded | 6a0d3ff5f984739ed67376b4 |
| DeltaNet | 42.2 | Not recorded | 6a0d3ff5f984739ed67376b4 |
| GLA | 38.8 | Not recorded | 6a0d3ff5f984739ed67376b4 |

### FLASH PD-SSM

Compress: 58. Result ID: 6a0d64d4fbc373592c052220.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d3ff5f984739ed67376b4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d3ff5f984739ed67376b4.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Mamba

Compress: 52.7. Result ID: 6a0d64d4fbc373592c05221d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d3ff5f984739ed67376b4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73611e209131cea5c0fe.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Transformer

Compress: 51.6. Result ID: 6a0d64d4fbc373592c05221c.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d3ff5f984739ed67376b4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78b71e209131cea7a00e.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### DeltaNet

Compress: 42.2. Result ID: 6a0d64d4fbc373592c05221f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d3ff5f984739ed67376b4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76981e209131cea6e083.md>)

[Implementation](<https://github.com/sustcsonglin/flash-linear-attention>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### GLA

Compress: 38.8. Result ID: 6a0d64d4fbc373592c05221e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d3ff5f984739ed67376b4.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

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