Nanda 10B Chat
Nanda 10B Chat is a Hindi-English model based on Llama 3 architecture. The collaboration trained it with a corpus containing 65 billion Hindi tokens. The official weights are hosted by MBZUAI, not under the current Inception42 repository path.
Restricted access
Check the model class
The card requires a custom model class and trust_remote_code when loading. It targets Hindi and English; it explicitly warns against assuming equal ability in other languages.Nanda 10B official MBZUAI model card
Test Hindi inputs as users write them
Practical guidance. Build evaluation examples from the actual documents and messages the application receives. Include mixed Hindi-English wording and domain terms. The size of the Hindi training corpus does not establish accuracy for a specific business vocabulary.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Restricted access |
| Access | Downloadable |
| Model identifiers | Llama-3-Nanda-10B-Chat |
| License | Llama 3 terms; gated repositoryNanda 10B official MBZUAI model card |
| Developers | MBZUAI Institute of Foundation Models, Inception and CerebrasNanda 10B official MBZUAI model card |
Related models
- Sherkala 8B Chat
Inception42 (G42) ยท Sherkala
Eight-billion-parameter Llama 3.1-based instruction model focused on Kazakh, with English, Russian and Turkish support.
Launch date not confirmedRestricted accessDownloadable