# Controllable Text Generation on AGNews (test)

Source: [Controllable Text Generation on AGNews (test)](<https://www.sota2.com/research/sota/controllable-text-generation-on-agnews-test>)

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

Data updated: 2026-02-26T05:54:21.652Z

Evaluation of Controllable Text Generation performance on the test split of the AG News dataset, measured using Output Perplexity (O-PPL), Metric Perplexity (M-PPL), F1 score, Diversity (Dist), and Style/Attribute Balance (S-BL).

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 | Controllable Text Generation |
| Dataset (including any stated split) | AGNews (test) |
| Metric | Output Perplexity (O-PPL) |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 12 of 12 |
| 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

- [Output Perplexity (O-PPL)](<https://www.sota2.com/research/sota/controllable-text-generation-on-agnews-test.md?metric=Output%20Perplexity%20(O-PPL)>): Lower is better; unit: SCALAR.

- [Metric Perplexity (M-PPL)](<https://www.sota2.com/research/sota/controllable-text-generation-on-agnews-test.md?metric=Metric%20Perplexity%20(M-PPL)>): Lower is better; unit: SCALAR.

- [F1 Score](<https://www.sota2.com/research/sota/controllable-text-generation-on-agnews-test.md?metric=F1%20Score>): Higher is better; unit: PERCENT.

- [Diversity (Dist)](<https://www.sota2.com/research/sota/controllable-text-generation-on-agnews-test.md?metric=Diversity%20(Dist)>): Higher is better; unit: PERCENT.

- [Style/Attribute Balance (S-BL)](<https://www.sota2.com/research/sota/controllable-text-generation-on-agnews-test.md?metric=Style%2FAttribute%20Balance%20(S-BL)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Output Perplexity (O-PPL) | Result date | Report paper |
| --- | --- | --- | --- |
| GPT2(raw) | 16.94 | Not recorded | 68fa6fa11e209131cea46ca8 |
| PF | 20.27 | Not recorded | 68fa6fa11e209131cea46ca8 |
| Ctr-PF | 20.41 | Not recorded | 68fa6fa11e209131cea46ca8 |
| GPT2 | 22.22 | Not recorded | 68fa6fa11e209131cea46ca8 |
| Ground Truth | 31.04 | Not recorded | 68fa6fa11e209131cea46ca8 |
| PT(noise)+PL | 32.36 | Not recorded | 68fa6fa11e209131cea46ca8 |
| PT(select)+PL | 33.23 | Not recorded | 68fa6fa11e209131cea46ca8 |
| DuNST | 34.73 | Not recorded | 68fa6fa11e209131cea46ca8 |
| T5 | 48.33 | Not recorded | 68fa6fa11e209131cea46ca8 |
| UniLM | 55.79 | Not recorded | 68fa6fa11e209131cea46ca8 |
| PT | 57.4 | Not recorded | 68fa6fa11e209131cea46ca8 |
| PT(noise) | 58.59 | Not recorded | 68fa6fa11e209131cea46ca8 |

### GPT2(raw)

Output Perplexity (O-PPL): 16.94. Result ID: 6990e52d3a6c314849041848.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

Recorded evaluation configuration: {"Method Category":"Zero-shot PLM"}. This record does not assert independent reproduction.

### PF

Output Perplexity (O-PPL): 20.27. Result ID: 6990e52d3a6c31484904184c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa690f1e209131cea1fe0e.md>)

Recorded evaluation configuration: {"Method Category":"Lightweight method"}. This record does not assert independent reproduction.

### Ctr-PF

Output Perplexity (O-PPL): 20.41. Result ID: 6990e52d3a6c31484904184d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6bdd1e209131cea30afd.md>)

Recorded evaluation configuration: {"Method Category":"Lightweight method"}. This record does not assert independent reproduction.

### GPT2

Output Perplexity (O-PPL): 22.22. Result ID: 6990e52d3a6c314849041849.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

Recorded evaluation configuration: {"Method Category":"Finetuned PLM"}. This record does not assert independent reproduction.

### Ground Truth

Output Perplexity (O-PPL): 31.04. Result ID: 6990e52d3a6c314849041847.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

Recorded evaluation configuration: {"Method Category":"Reference"}. This record does not assert independent reproduction.

### PT(noise)+PL

Output Perplexity (O-PPL): 32.36. Result ID: 6990e52d3a6c314849041850.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa688b1e209131cea1cd36.md>)

Recorded evaluation configuration: {"Method Category":"Self-Training with PLM","Backbone PLM":"UniLM","Pseudo Labeler":"BERT-base"}. This record does not assert independent reproduction.

### PT(select)+PL

Output Perplexity (O-PPL): 33.23. Result ID: 6990e52d3a6c314849041851.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

Recorded evaluation configuration: {"Method Category":"Self-Training with PLM","Backbone PLM":"UniLM","Pseudo Labeler":"BERT-base"}. This record does not assert independent reproduction.

### DuNST

Output Perplexity (O-PPL): 34.73. Result ID: 6990e52d3a6c314849041852.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Implementation](<https://github.com/peterfengyx/DuNST>)

Recorded evaluation configuration: {"Method Category":"Our method"}. This record does not assert independent reproduction.

### T5

Output Perplexity (O-PPL): 48.33. Result ID: 6990e52d3a6c31484904184b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa709e1e209131cea4c470.md>)

[Implementation](<https://github.com/google-research/text-to-text-transfer-transformer>)

Recorded evaluation configuration: {"Method Category":"Finetuned PLM"}. This record does not assert independent reproduction.

### UniLM

Output Perplexity (O-PPL): 55.79. Result ID: 6990e52d3a6c31484904184a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66831e209131cea10caa.md>)

[Implementation](<https://github.com/microsoft/unilm>)

Recorded evaluation configuration: {"Method Category":"Finetuned PLM"}. This record does not assert independent reproduction.

### PT

Output Perplexity (O-PPL): 57.4. Result ID: 6990e52d3a6c31484904184e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

Recorded evaluation configuration: {"Method Category":"Self-Training with PLM","Backbone PLM":"UniLM"}. This record does not assert independent reproduction.

### PT(noise)

Output Perplexity (O-PPL): 58.59. Result ID: 6990e52d3a6c31484904184f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6fa11e209131cea46ca8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa688b1e209131cea1cd36.md>)

Recorded evaluation configuration: {"Method Category":"Self-Training with PLM","Backbone PLM":"UniLM"}. This record does not assert independent reproduction.

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