# Language Modeling on WikiText-2 raw-v1 (test)

Source: [Language Modeling on WikiText-2 raw-v1 (test)](<https://www.sota2.com/research/sota/language-modeling-on-wikitext-2-raw-v1-test>)

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

Data updated: 2026-05-18T09:28:33.819Z

Evaluation of language modeling performance using the test split of the raw version 1 of the WikiText-2 dataset, focusing on Delta Perplexity (ΔPPL) and Information/Reduction (I/R) 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) | WikiText-2 raw-v1 (test) |
| Metric | Delta PPL (n=1) |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 8 of 8 |
| 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 (n=1)](<https://www.sota2.com/research/sota/language-modeling-on-wikitext-2-raw-v1-test.md?metric=Delta%20PPL%20(n%3D1)>): Lower is better; unit: SCALAR.

- [Delta PPL (n=3)](<https://www.sota2.com/research/sota/language-modeling-on-wikitext-2-raw-v1-test.md?metric=Delta%20PPL%20(n%3D3)>): Lower is better; unit: SCALAR.

- [Delta PPL (n=5)](<https://www.sota2.com/research/sota/language-modeling-on-wikitext-2-raw-v1-test.md?metric=Delta%20PPL%20(n%3D5)>): Lower is better; unit: SCALAR.

- [Information/Reduction (n=1)](<https://www.sota2.com/research/sota/language-modeling-on-wikitext-2-raw-v1-test.md?metric=Information%2FReduction%20(n%3D1)>): Higher is better; unit: PERCENT.

- [Information/Reduction (n=3)](<https://www.sota2.com/research/sota/language-modeling-on-wikitext-2-raw-v1-test.md?metric=Information%2FReduction%20(n%3D3)>): Higher is better; unit: PERCENT.

- [Information/Reduction (n=5)](<https://www.sota2.com/research/sota/language-modeling-on-wikitext-2-raw-v1-test.md?metric=Information%2FReduction%20(n%3D5)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Delta PPL (n=1) | Result date | Report paper |
| --- | --- | --- | --- |
| Interchange (clustered) | 2.5 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |
| SLEB-iterative | 2.5 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |
| BI-guided | 2.5 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |
| Interchange | 11 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |
| BI-guided | 11 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |
| Replacement | 11.3 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |
| SLEB-iterative | 11.6 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |
| Replacement | 16.7 | Not recorded | 6a0a9d3adbc836b87d82b5e8 |

### Interchange (clustered)

Delta PPL (n=1): 2.5. Result ID: 6a0adaf247f34ce924f14df3.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Precision":"bf16","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

### SLEB-iterative

Delta PPL (n=1): 2.5. Result ID: 6a0adaf247f34ce924f14df5.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76261e209131cea6b833.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Precision":"bf16","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

### BI-guided

Delta PPL (n=1): 2.5. Result ID: 6a0adaf247f34ce924f14df6.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

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

Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Precision":"bf16","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

### Interchange

Delta PPL (n=1): 11. Result ID: 6a0adaf247f34ce924f14df7.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

Recorded evaluation configuration: {"Backbone":"Pythia-1.4B","Precision":"fp32","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

### BI-guided

Delta PPL (n=1): 11. Result ID: 6a0adaf247f34ce924f14dfa.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

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

Recorded evaluation configuration: {"Backbone":"Pythia-1.4B","Precision":"fp32","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

### Replacement

Delta PPL (n=1): 11.3. Result ID: 6a0adaf247f34ce924f14df4.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Precision":"bf16","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

### SLEB-iterative

Delta PPL (n=1): 11.6. Result ID: 6a0adaf247f34ce924f14df9.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76261e209131cea6b833.md>)

Recorded evaluation configuration: {"Backbone":"Pythia-1.4B","Precision":"fp32","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

### Replacement

Delta PPL (n=1): 16.7. Result ID: 6a0adaf247f34ce924f14df8.

[Reporting paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0a9d3adbc836b87d82b5e8.md>)

Recorded evaluation configuration: {"Backbone":"Pythia-1.4B","Precision":"fp32","Words":"5K","Window":"512","Stride":"256"}. This record does not assert independent reproduction.

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