OLA founder and chairman Bhavish Aggarwal announced the Krutrim AI Lab on Tuesday. The AI-centric research lab will be the central hub for all future versions of Ola Krutrim (often called krutrim). In addition, several new India-centric open source artificial intelligence (AI) models have been released, with the main focus on the Krutrim-2 AI model. Agwal also announced a investment in Rs. Krutrim’s Rs 2 million crore and made an investment of Rs. There will be 100 million from next year to a year.
Krutrim AI lab and new model release
In a post by X (formerly known as Twitter), Aggarwal stressed that his AI company has been working to develop new AI innovations since its inception. Chairman Ola demonstrated what Krutrim has achieved last year and released several new open source AI models. He added: “Our focus is on developing AI for India to make AI better in terms of language, data scarcity, cultural background, etc.”
The biggest highlight of the released models is Krutrim-2, a 12 billion parameter model that is the successor to the Krutrim-1 AI model. This is an intensive transformer model built on the Mistral-Nemo-12b-12b-Instruct structure. The native multilingual model can respond in English and 22 Indian regional languages and supports a context window of 1,28,000 tokens. The model is currently available through a list of embracing faces with the Krutrim community for academic and research-oriented usage.
Krutrim also announced the release of Chitrarth-1, a multilingual visual language model (VLM) based on Krutrim-7B. It uses a siglip vision encoder to extract and process visual information. The company said that AI models are trained in multiple languages of image and text data and can span the following: Assam, Bengali, Gujarati, Kannada, Malayalam, Malayalam, Malayalam, Works in nine Indian regional languages including Rayalam, Marathi, Odia, Tamil and Telugu. It also knows English.
For AI-based voice translation, the company released the DHWANI-1 automatic speech recognition (ASR) model. It supports translation between the indicative language and English. Supported languages are English, Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Tamil and Telugu. It is accompanied by krutrim translation text translation model.
Aggarwal also announced the release of Vyakyarth-1 Indian Insert, a multilingual sentence converter model designed to conduct semantic text similarity, search, clustering and classification across 100 languages. He added that the model may be useful for translingual natural language processing (NLP) applications.
Additionally, Aggarwal stressed that the company has also developed a new benchmark dubbing called Bharatench, which measures the performance of AI models on INDIC. He claims that the benchmarks tested India’s unique linguistic and cultural disadvantages, which were not seen in other tests.
“We are not close to global benchmarks, but have made good progress in 1 year. By opening up our model, we hope that the entire Indian AI community will work together to create a world-class Indian AI ecosystem,” Aggarwal said.
Krutrim has also partnered with NVIDIA to deploy the technology giant’s GB200 GPU for AI-ai-ai-abable workloads. The infrastructure will be launched in March.
