KyutaiARC Encoder
ARC8 Encoder Multi
ARC8 Encoder Multi compresses text into representations that a language-model decoder can use. Its pooling factor is eight. It was trained with both Llama 3.1 8B and Mistral 7B decoders.
Research release
The encoder does not answer by itself
Its output is an encoded representation, not a finished reply. Downstream use needs a compatible decoder and may require further tuning. Even the Mistral-named variant retains Llama ancestry in its encoder; the name describes the target decoder.ARC8_Encoder_multi: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/ARC8_Encoder_multi |
| Encoder ancestry | Llama 3.2 3B backbone.ARC8_Encoder_multi: Kyutai model card |
| Terms | CC BY 4.0 plus the applicable Llama terms.ARC8_Encoder_multi: Kyutai model card |
Related models
A Moshika speech front end that receives retrieved text while an audio conversation continues.
Launch date not confirmedResearch releaseDownloadable