Indian AI lab Sarvam’s new models are a major bet on the viability of open-source AI

Indian AI startup Sarvam has launched a new generation of open-source models at the India AI Impact Summit, including 30B and 105B parameter LLMs using mixture-of-experts architecture, plus text-to-speech, speech-to-text, and vision models. Both large models were trained from scratch — the 30B on ~16 trillion tokens, the 105B Show more