18 Dec 2023
The company claims that Krutrim outperforms Chat GPT4 and Meta’s Llama 2 chat in Indian languages regarding computation and speed.
On Friday, Ola Founder Bhavish Aggarwal announced the public launch of India’s own Large Language Model (LLM) Krutrim in January 2024. This model will offer generative support for 10 Indian languages and be able to take inputs for 22 languages.
The company claims that Krutrim’s current model outperforms OpenAI’s GPT4 in terms of computation and speed, and Meta’s Llama 2 chat. Although it couldn’t beat Google’s Bard and Gemini. Krutrim can operate on voice and text commands. An advanced Krutrim Pro model will be launched in the next quarter of 2024. This Pro software will offer task, vision, and speech execution abilities.
The developers used more than 2 trillion tokens of data to train Krutrim in Indian languages. A token is a basic data unit like a text or code an LLM uses to process and generate language. The company used an Indian-made tokenizer to build Krutrim.