# Natural Language Generation on Math

Source: [Natural Language Generation on Math](<https://www.sota2.com/research/sota/natural-language-generation-on-math>)

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

Data updated: 2026-02-27T09:24:20.396Z

Evaluation of Natural Language Generation performance on the Math dataset, assessed using metrics such as BLEU, ROUGE, and BERT-F.

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 | Natural Language Generation |
| Dataset (including any stated split) | Math |
| Metric | BLEU |
| Unit | PERCENT |
| Direction | Higher 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

- [BLEU](<https://www.sota2.com/research/sota/natural-language-generation-on-math.md?metric=BLEU>): Higher is better; unit: PERCENT.

- [ROUGE](<https://www.sota2.com/research/sota/natural-language-generation-on-math.md?metric=ROUGE>): Higher is better; unit: PERCENT.

- [BERT-F](<https://www.sota2.com/research/sota/natural-language-generation-on-math.md?metric=BERT-F>): Higher is better; unit: PERCENT.

## Recorded results

| Method | BLEU | Result date | Report paper |
| --- | --- | --- | --- |
| ADORE | 38.8 | Not recorded | 68fa73e61e209131cea5ec78 |
| Full Attention | 38.6 | Not recorded | 68fa73e61e209131cea5ec78 |
| H2O(Rebuilt) | 34.7 | Not recorded | 68fa73e61e209131cea5ec78 |
| H2O | 33.3 | Not recorded | 68fa73e61e209131cea5ec78 |
| Strided Attention | 33 | Not recorded | 68fa73e61e209131cea5ec78 |
| StreamingLLM | 32.9 | Not recorded | 68fa73e61e209131cea5ec78 |
| KV Compression | 32.2 | Not recorded | 68fa73e61e209131cea5ec78 |
| Window Attention | 30.3 | Not recorded | 68fa73e61e209131cea5ec78 |

### ADORE

BLEU: 38.8. Result ID: 69a162b935dfcabb00e6b6de.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

[Implementation](<https://github.com/WHUIR/ADORE>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000","KV cache size":"192"}. This record does not assert independent reproduction.

### Full Attention

BLEU: 38.6. Result ID: 69a162b935dfcabb00e6b6d7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000"}. This record does not assert independent reproduction.

### H2O(Rebuilt)

BLEU: 34.7. Result ID: 69a162b935dfcabb00e6b6dd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000","KV cache size":"192"}. This record does not assert independent reproduction.

### H2O

BLEU: 33.3. Result ID: 69a162b935dfcabb00e6b6dc.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718c1e209131cea51ead.md>)

[Implementation](<https://github.com/FMInference/H2O>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000","KV cache size":"192"}. This record does not assert independent reproduction.

### Strided Attention

BLEU: 33. Result ID: 69a162b935dfcabb00e6b6d9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65aa1e209131cea0bc26.md>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000","KV cache size":"192"}. This record does not assert independent reproduction.

### StreamingLLM

BLEU: 32.9. Result ID: 69a162b935dfcabb00e6b6db.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72d71e209131cea58d79.md>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000","KV cache size":"192"}. This record does not assert independent reproduction.

### KV Compression

BLEU: 32.2. Result ID: 69a162b935dfcabb00e6b6da.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70ea1e209131cea4df67.md>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000","KV cache size":"192"}. This record does not assert independent reproduction.

### Window Attention

BLEU: 30.3. Result ID: 69a162b935dfcabb00e6b6d8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73e61e209131cea5ec78.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f501e209131cea44f4a.md>)

[Implementation](<https://github.com/SHI-Labs/Neighborhood-Attention-Transformer>)

Recorded evaluation configuration: {"Backbone":"Llama-2 7B","Context length":"4,000","KV cache size":"192"}. This record does not assert independent reproduction.

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