# Language Modeling on Code, Science, and Fiction domains Equal-Weight Evaluation

Source: [Language Modeling on Code, Science, and Fiction domains Equal-Weight Evaluation](<https://www.sota2.com/research/sota/language-modeling-on-code-science-and-fiction-domains-equal-weight-evaluation>)

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

Data updated: 2026-03-25T14:26:51.435Z

Evaluation of Language Modeling performance across Code, Science, and Fiction domains, using an Equal-Weight scoring methodology.

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) | Code, Science, and Fiction domains Equal-Weight Evaluation |
| Metric | EW Loss |
| 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

- [EW Loss](<https://www.sota2.com/research/sota/language-modeling-on-code-science-and-fiction-domains-equal-weight-evaluation.md?metric=EW%20Loss>): Lower is better; unit: SCALAR.

- [Relative Change vs Best Spec.](<https://www.sota2.com/research/sota/language-modeling-on-code-science-and-fiction-domains-equal-weight-evaluation.md?metric=Relative%20Change%20vs%20Best%20Spec.>): Higher is better; unit: SCALAR.

- [Relative Change vs Base](<https://www.sota2.com/research/sota/language-modeling-on-code-science-and-fiction-domains-equal-weight-evaluation.md?metric=Relative%20Change%20vs%20Base>): Higher is better; unit: SCALAR.

- [Std Dev (EW Loss)](<https://www.sota2.com/research/sota/language-modeling-on-code-science-and-fiction-domains-equal-weight-evaluation.md?metric=Std%20Dev%20(EW%20Loss)>): Lower is better; unit: SCALAR.

## Recorded results

| Method | EW Loss | Result date | Report paper |
| --- | --- | --- | --- |
| KALAVAI (MoE) | 2.09 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Monolithic baseline | 2.097 | Not recorded | 69c36c040636e5bf4afc64a8 |
| KALAVAI (MoE) | 2.118 | Not recorded | 69c36c040636e5bf4afc64a8 |
| KALAVAI (MoE) | 2.218 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Monolithic baseline | 2.229 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Best specialist | 2.259 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Best specialist | 2.266 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Base model | 2.32 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Best specialist | 2.404 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Base model | 2.474 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Weight averaging | 2.486 | Not recorded | 69c36c040636e5bf4afc64a8 |
| Base model | 2.651 | Not recorded | 69c36c040636e5bf4afc64a8 |

### KALAVAI (MoE)

EW Loss: 2.09. Result ID: 69c3f03247f34ce924adc4a3.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-1B","Seeds":"3"}. This record does not assert independent reproduction.

### Monolithic baseline

EW Loss: 2.097. Result ID: 69c3f03247f34ce924adc4a2.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-1B","Seeds":"1"}. This record does not assert independent reproduction.

### KALAVAI (MoE)

EW Loss: 2.118. Result ID: 69c3f03247f34ce924adc4a6.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-6.9B","Seeds":"3"}. This record does not assert independent reproduction.

### KALAVAI (MoE)

EW Loss: 2.218. Result ID: 69c3f03247f34ce924adc49f.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-410M","Seeds":"3"}. This record does not assert independent reproduction.

### Monolithic baseline

EW Loss: 2.229. Result ID: 69c3f03247f34ce924adc49e.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-410M","Seeds":"3"}. This record does not assert independent reproduction.

### Best specialist

EW Loss: 2.259. Result ID: 69c3f03247f34ce924adc4a1.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-1B","Seeds":"1"}. This record does not assert independent reproduction.

### Best specialist

EW Loss: 2.266. Result ID: 69c3f03247f34ce924adc4a5.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-6.9B","Seeds":"1"}. This record does not assert independent reproduction.

### Base model

EW Loss: 2.32. Result ID: 69c3f03247f34ce924adc4a4.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-6.9B","Seeds":"—"}. This record does not assert independent reproduction.

### Best specialist

EW Loss: 2.404. Result ID: 69c3f03247f34ce924adc49c.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-410M","Seeds":"3"}. This record does not assert independent reproduction.

### Base model

EW Loss: 2.474. Result ID: 69c3f03247f34ce924adc4a0.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-1B","Seeds":"—"}. This record does not assert independent reproduction.

### Weight averaging

EW Loss: 2.486. Result ID: 69c3f03247f34ce924adc49d.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-410M","Seeds":"3"}. This record does not assert independent reproduction.

### Base model

EW Loss: 2.651. Result ID: 69c3f03247f34ce924adc49b.

[Reporting paper](<https://www.sota2.com/research/paper/69c36c040636e5bf4afc64a8.md>)

Recorded evaluation configuration: {"Scale":"Pythia-410M","Seeds":"—"}. This record does not assert independent reproduction.

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