Sarvam AI Launches Open-Weights Sarvam-2B Model and Suite of AI Products for Indian Languages

Sarvam AI Launches Open-Weights Sarvam-2B Model and Suite of AI Products for Indian Languages

Sarvam AI, a Bengaluru-based startup, officially launched its full-stack generative AI platform today, introducing a range of products and services designed to cater to India’s diverse linguistic. The company, which raised $41 million in Series A funding last December, aims to position itself as a leader in developing AI solutions tailored for Indian needs.

Headlining the launch is Sarvam-2B, a 2 billion parameter large language model (LLM) trained from scratch with a significant focus on Indic languages. According to the company, Sarvam-2B supports 10 Indian languages: Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Oriya, Punjabi, Tamil, and Telugu.

The company claims that Sarvam-2B, despite its relatively small size, outperforms larger models like Llama 3.1-8B in translation tasks. This early version of the model has been trained on 2 trillion tokens out of a planned 4 trillion, with half of the data comprising Indic language content. Sarvam-2B is now available on Hugging Face for non-commercial and research purposes. Commercial use requires explicit permission from Sarvam AI.

In addition to Sarvam-2B, the company introduced several other AI models and products:

  1. Bulbul 1.0: A text-to-speech model supporting 10 languages with six voice options.
  2. Saaras 1.0 and Saarika: Speech-to-text models capable of handling 10 Indian languages with automatic language identification.
  3. Mayura 1.0: A translation API designed to handle the complexities of Indian languages, supporting formal and colloquial translations.
  4. Shuka v1: An encoder-decoder model combining Sarvam's audio encoder with Meta's Llama-3 for understanding audio in Indic languages.

Sarvam AI is also launching enterprise-focused products, including SarvamAgents, voice-enabled AI agents for customer interactions, and A1, a generative AI workbench for legal professionals.

To support its AI infrastructure, Sarvam has partnered with technology providers including Yotta, Nvidia, Microsoft Azure, Google Cloud Platform, and Exotel. The company's models will be powered by Nvidia's H100 chips, according to Vishal Dhupar, Nvidia's South Asia managing director.

Sarvam AI co-founder Vivek Raghavan emphasized the company's strategic focus on building AI solutions tailored for India's specific needs, particularly in language processing and voice-first interfaces. The startup hopes to collaborate with Indian companies across various sectors to develop customized AI models.

Chris McKay is the founder and chief editor of Maginative. His thought leadership in AI literacy and strategic AI adoption has been recognized by top academic institutions, media, and global brands.

Let’s stay in touch. Get the latest AI news from Maginative in your inbox.

Subscribe