Sarvam
Indian full-stack sovereign AI lab building foundation models trained from scratch for 22 Indic languages. Model family includes Sarvam-30B and Sarvam-105B (MoE, Apache 2.0), plus Saaras v3 ASR, Bulbul TTS, Mayura translation, and a 3B state-space vision-language model. Inference via vLLM; trained on NVIDIA NeMo. Selected under India's IndiaAI Mission.
