# Online Language Modeling on PG-19 (Whole Book)

Source: [Online Language Modeling on PG-19 (Whole Book)](<https://www.sota2.com/research/sota/online-language-modeling-on-pg-19-whole-book>)

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

Data updated: 2026-02-26T05:55:12.146Z

Evaluation of online language modeling performance on the full PG-19 corpus, measured by Perplexity at various context lengths (50K, 100K, 150K, 200K tokens).

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 | Online Language Modeling |
| Dataset (including any stated split) | PG-19 (Whole Book) |
| Metric | PPL @ 50K |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 17 of 17 |
| 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

- [PPL @ 50K](<https://www.sota2.com/research/sota/online-language-modeling-on-pg-19-whole-book.md?metric=PPL%20%40%2050K>): Lower is better; unit: SCALAR.

- [PPL @ 100K](<https://www.sota2.com/research/sota/online-language-modeling-on-pg-19-whole-book.md?metric=PPL%20%40%20100K>): Lower is better; unit: SCALAR.

- [PPL @ 150K](<https://www.sota2.com/research/sota/online-language-modeling-on-pg-19-whole-book.md?metric=PPL%20%40%20150K>): Lower is better; unit: SCALAR.

- [PPL @ 200K](<https://www.sota2.com/research/sota/online-language-modeling-on-pg-19-whole-book.md?metric=PPL%20%40%20200K>): Lower is better; unit: SCALAR.

## Recorded results

| Method | PPL @ 50K | Result date | Report paper |
| --- | --- | --- | --- |
| TTT w/ Locas-MLP + NL-SVD | 18.87 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ Locas-MLP + BP | 18.89 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ TempLoRA | 18.94 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Cxt. Trunc. (4K) | 18.98 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Long Cxt. Attn. (16K) | 18.98 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Cxt. Trunc. (2K) | 19.35 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ Locas-GLU | 20 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ TempLoRA | 20.03 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Cxt. Trunc. (2K) | 20.6 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ Locas-GLU | 23.65 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ TempLoRA | 23.74 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Long Cxt. Attn. (16K) | 25.24 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ Locas-GLU | 26.58 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| TTT w/ TempLoRA | 26.71 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Cxt. Trunc. (2K) | 27.49 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Long Cxt. Attn. (16K) | 31.45 | Not recorded | 69857568c97a0b3e3b5d5b0a |
| Cxt. Trunc. (2K) | 31.59 | Not recorded | 69857568c97a0b3e3b5d5b0a |

### TTT w/ Locas-MLP + NL-SVD

PPL @ 50K: 18.87. Result ID: 699649a55698665ec48dcf1d.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"0.6B MLP-FFN LM","Compression":"NL-SVD","Extra #Params":"4.2M","Relative Time":"12.7x"}. This record does not assert independent reproduction.

### TTT w/ Locas-MLP + BP

PPL @ 50K: 18.89. Result ID: 699649a55698665ec48dcf1e.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"0.6B MLP-FFN LM","Optimization":"Back-propagation (BP)","Extra #Params":"4.2M","Relative Time":"1.7x"}. This record does not assert independent reproduction.

### TTT w/ TempLoRA

PPL @ 50K: 18.94. Result ID: 699649a55698665ec48dcf1c.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74c11e209131cea638ed.md>)

Recorded evaluation configuration: {"Backbone":"0.6B MLP-FFN LM","Extra #Params":"41.9M","Relative Time":"4.7x"}. This record does not assert independent reproduction.

### Cxt. Trunc. (4K)

PPL @ 50K: 18.98. Result ID: 699649a55698665ec48dcf1b.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"0.6B MLP-FFN LM","Extra #Params":"0","Relative Time":"1.4x"}. This record does not assert independent reproduction.

### Long Cxt. Attn. (16K)

PPL @ 50K: 18.98. Result ID: 699649a55698665ec48dcf24.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B-Base","Extra #Params":"0","Relative Time":"18.6x"}. This record does not assert independent reproduction.

### Cxt. Trunc. (2K)

PPL @ 50K: 19.35. Result ID: 699649a55698665ec48dcf1a.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"0.6B MLP-FFN LM","Extra #Params":"0","Relative Time":"1x"}. This record does not assert independent reproduction.

### TTT w/ Locas-GLU

PPL @ 50K: 20. Result ID: 699649a55698665ec48dcf26.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B-Base","Extra #Params":"11.0M","Relative Time":"1.8x"}. This record does not assert independent reproduction.

### TTT w/ TempLoRA

PPL @ 50K: 20.03. Result ID: 699649a55698665ec48dcf25.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74c11e209131cea638ed.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B-Base","Extra #Params":"73.4M","Relative Time":"4.7x"}. This record does not assert independent reproduction.

### Cxt. Trunc. (2K)

PPL @ 50K: 20.6. Result ID: 699649a55698665ec48dcf23.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B-Base","Extra #Params":"0","Relative Time":"1x"}. This record does not assert independent reproduction.

### TTT w/ Locas-GLU

PPL @ 50K: 23.65. Result ID: 699649a55698665ec48dcf2a.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B(-Instruct)","Extra #Params":"11.0M","Relative Time":"1.8x"}. This record does not assert independent reproduction.

### TTT w/ TempLoRA

PPL @ 50K: 23.74. Result ID: 699649a55698665ec48dcf29.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74c11e209131cea638ed.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B(-Instruct)","Extra #Params":"73.4M","Relative Time":"4.7x"}. This record does not assert independent reproduction.

### Long Cxt. Attn. (16K)

PPL @ 50K: 25.24. Result ID: 699649a55698665ec48dcf20.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B-Base","Extra #Params":"0","Relative Time":"13.9x"}. This record does not assert independent reproduction.

### TTT w/ Locas-GLU

PPL @ 50K: 26.58. Result ID: 699649a55698665ec48dcf22.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B-Base","Extra #Params":"5.5M","Relative Time":"1.4x"}. This record does not assert independent reproduction.

### TTT w/ TempLoRA

PPL @ 50K: 26.71. Result ID: 699649a55698665ec48dcf21.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74c11e209131cea638ed.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B-Base","Extra #Params":"36.7M","Relative Time":"4.3x"}. This record does not assert independent reproduction.

### Cxt. Trunc. (2K)

PPL @ 50K: 27.49. Result ID: 699649a55698665ec48dcf1f.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-0.6B-Base","Extra #Params":"0","Relative Time":"1x"}. This record does not assert independent reproduction.

### Long Cxt. Attn. (16K)

PPL @ 50K: 31.45. Result ID: 699649a55698665ec48dcf28.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B(-Instruct)","Extra #Params":"0","Relative Time":"18.6x"}. This record does not assert independent reproduction.

### Cxt. Trunc. (2K)

PPL @ 50K: 31.59. Result ID: 699649a55698665ec48dcf27.

[Reporting paper](<https://www.sota2.com/research/paper/69857568c97a0b3e3b5d5b0a.md>)

Recorded evaluation configuration: {"Backbone":"Qwen3-1.7B(-Instruct)","Extra #Params":"0","Relative Time":"1x"}. This record does not assert independent reproduction.

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