Helium 1 2B
Helium 1 is a two-billion-parameter base language model for the 24 official EU languages. It expands the six-language preview and remains a text-completion foundation rather than an instruction-tuned assistant.
Research release
The final model combines trained variants
Kyutai produced specialized checkpoints from different parts of its filtered web data. The final Helium 1 release combines the main run with selected specialized models through weighted parameter averaging.Helium 1: modular training and model combinations
The card recommends further alignment
For instruction-following applications, the card calls for supervised fine-tuning, preference training or a related method. A small parameter count does not remove the need to check output quality in each language.helium-1-2b: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/helium-1-2b |
| Architecture | 28 layers, hidden width 2,048 and a 4,096-token context.helium-1-2b: Kyutai model card |
| Terms | CC BY-SA 4.0, with Gemma Terms of Use because training distilled Gemma 2.helium-1-2b: Kyutai model card |
Related models
- Dactory language classifiers
Kyutai · Dactory
A set of fastText classifiers that filters Common Crawl text in the 24 official EU languages.
Launch date not confirmedResearch releaseDownloadable - Helium 1 2B Books
Kyutai · Helium
A 2B Helium base checkpoint trained on the textbook-oriented part of its filtered data.
Launch date not confirmedResearch releaseDownloadable - Helium 1 Preview 2B
Kyutai · Helium
A 2B base language model for English, French, German, Italian, Portuguese and Spanish.
Launch date not confirmedResearch releaseDownloadable