← Catalog
Lara Mini Base
AvailableTranslatedOpen weights
Translated's first general-purpose LLM: a research base model trained from scratch on 6T tokens across 45 languages (all 24 official EU languages). Sparse MoE, 7.2B total / 0.6B active (4 of 128 routed experts plus a shared expert), hybrid Gated DeltaNet, sliding-window and global attention, 131,072-token context. Gated release on Hugging Face (translated/Lara-4-Mini-Base) under CC BY-NC-SA 4.0; no instruction or safety tuning.
Specifications
- License
- Open weights · CC BY-NC-SA 4.0
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 7.2B · 0.6B active
- Context window
- 131K tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- —
- Modalities
- Text
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.