# Language Modeling on GPT-2

Source: [Language Modeling on GPT-2](<https://www.sota2.com/research/sota/language-modeling-on-gpt-2>)

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

Data updated: 2026-04-28T07:00:07.615Z

Evaluation of language modeling performance on the general GPT-2 dataset, measured by Delta PPL and L-N/S metrics.

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 | Language Modeling |
| Dataset (including any stated split) | GPT-2 |
| Metric | Delta PPL |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 32 of 32 |
| 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

- [Delta PPL](<https://www.sota2.com/research/sota/language-modeling-on-gpt-2.md?metric=Delta%20PPL>): Lower is better; unit: SCALAR.

- [L-N/S](<https://www.sota2.com/research/sota/language-modeling-on-gpt-2.md?metric=L-N%2FS>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Delta PPL | Result date | Report paper |
| --- | --- | --- | --- |
| Random | 0.44 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 0.5 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 0.62 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 0.73 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 0.75 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 0.84 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 1.04 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Random | 1.06 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Random | 1.06 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Random | 1.51 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 1.68 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 1.92 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 1.98 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Random | 2.18 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Random | 2.36 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 2.66 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 2.67 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 2.96 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 3.41 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Random | 3.79 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 4.89 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 5.55 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Random | 5.62 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 6.04 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Spectral | 9.65 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 13.49 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 19.49 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 21.2 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Last-N | 25.06 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 26.8 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 33.36 | Not recorded | 69f03f0755a3c32af89f9f94 |
| Worst | 229.15 | Not recorded | 69f03f0755a3c32af89f9f94 |

### Random

Delta PPL: 0.44. Result ID: 69f05a1847f34ce924d5a4b7.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"1"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 0.5. Result ID: 69f05a1847f34ce924d5a4b5.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"1"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 0.62. Result ID: 69f05a1847f34ce924d5a4b8.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"1"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 0.73. Result ID: 69f05a1847f34ce924d5a4a8.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"1"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 0.75. Result ID: 69f05a1847f34ce924d5a4b6.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"1"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 0.84. Result ID: 69f05a1847f34ce924d5a4a5.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"1"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 1.04. Result ID: 69f05a1847f34ce924d5a4b9.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"2"}. This record does not assert independent reproduction.

### Random

Delta PPL: 1.06. Result ID: 69f05a1847f34ce924d5a4a7.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"1"}. This record does not assert independent reproduction.

### Random

Delta PPL: 1.06. Result ID: 69f05a1847f34ce924d5a4bb.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"2"}. This record does not assert independent reproduction.

### Random

Delta PPL: 1.51. Result ID: 69f05a1847f34ce924d5a4bf.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"3"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 1.68. Result ID: 69f05a1847f34ce924d5a4bd.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"3"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 1.92. Result ID: 69f05a1847f34ce924d5a4ba.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"2"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 1.98. Result ID: 69f05a1847f34ce924d5a4a6.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"1"}. This record does not assert independent reproduction.

### Random

Delta PPL: 2.18. Result ID: 69f05a1847f34ce924d5a4c3.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"4"}. This record does not assert independent reproduction.

### Random

Delta PPL: 2.36. Result ID: 69f05a1847f34ce924d5a4ab.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"2"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 2.66. Result ID: 69f05a1847f34ce924d5a4ac.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"2"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 2.67. Result ID: 69f05a1847f34ce924d5a4a9.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"2"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 2.96. Result ID: 69f05a1847f34ce924d5a4c1.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"4"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 3.41. Result ID: 69f05a1847f34ce924d5a4be.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"3"}. This record does not assert independent reproduction.

### Random

Delta PPL: 3.79. Result ID: 69f05a1847f34ce924d5a4af.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"3"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 4.89. Result ID: 69f05a1847f34ce924d5a4ad.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"3"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 5.55. Result ID: 69f05a1847f34ce924d5a4c2.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"4"}. This record does not assert independent reproduction.

### Random

Delta PPL: 5.62. Result ID: 69f05a1847f34ce924d5a4b3.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"4"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 6.04. Result ID: 69f05a1847f34ce924d5a4aa.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"2"}. This record does not assert independent reproduction.

### Spectral

Delta PPL: 9.65. Result ID: 69f05a1847f34ce924d5a4b1.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"4"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 13.49. Result ID: 69f05a1847f34ce924d5a4ae.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"3"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 19.49. Result ID: 69f05a1847f34ce924d5a4b0.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"3"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 21.2. Result ID: 69f05a1847f34ce924d5a4bc.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"2"}. This record does not assert independent reproduction.

### Last-N

Delta PPL: 25.06. Result ID: 69f05a1847f34ce924d5a4b2.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75391e209131cea668f7.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"4"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 26.8. Result ID: 69f05a1847f34ce924d5a4c0.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"3"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 33.36. Result ID: 69f05a1847f34ce924d5a4c4.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Large (36L)","k":"4"}. This record does not assert independent reproduction.

### Worst

Delta PPL: 229.15. Result ID: 69f05a1847f34ce924d5a4b4.

[Reporting paper](<https://www.sota2.com/research/paper/69f03f0755a3c32af89f9f94.md>)

Recorded evaluation configuration: {"Model":"Medium (24L)","k":"4"}. This record does not assert independent reproduction.

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