6 months ago
42
Vishnu Vardhan, founder of SML and Vizzhy, speaking at Cypher 2024, India’s biggest AI conference by AIM Media House, announced that SML is going to release India’s first multimodal multilingual ‘Hanooman’ in the first week of October.
“We will be releasing India’s first multilingual multimodal Hanooman around the first week of October. We started working on 22 Indian languages, and now it works in 100 languages, which has been built from scratch in India,” said Vardhan.
Vardhan talked about practical applications of Hanooman in various sectors like healthcare. He presented a case study involving Hanooman analysing X-ray images. He described how the model significantly enhances the efficiency of radiologists, enabling them to process more images in less time. “A radiologist is reporting 10 X-rays an hour. With this, they can do 100,” he said.
He added that with this, they just need to verify the report generated by the model and, in some instances, make a few tweaks.
Vardhan also touched on the inherent limitations of existing language models, which often rely heavily on English data. He said, “There are 30 trillion tokens available for English, but when it comes to Indian languages, hardly any data exists.”
The scarcity of data in Indian languages—only about 200 million tokens—presents a major hurdle in developing effective AI solutions for the region. To address this, Hanooman employs a unique approach, ensuring equal representation of various languages to facilitate a more inclusive and effective model.
He also spoke about the financial aspects of maintaining large AI models. Vardhan revealed that his largest model, which contains 660 billion parameters, requires 80 GPUs to run continuously, costing millions of dollars even when idle. “That’s why all the big companies that have built these kinds of models have invested millions of dollars,” he explained. He expressed concern that such investments are rarely directed towards Indian startups.
Hanooman AI Studio Coming Soon
Moreover, Vardhan announced plans to launch Hanooman AI Studio in the third week of October. He explained that this platform will empower developers to create AI-driven applications without requiring coding expertise. “AI Studio will help developers make applications around AI without the requirement of the need to code,” he said, indicating a significant shift in how developers can approach AI solution building.
“We have created a studio that features a million specific workflows. You can develop a fintech app, create a healthcare agent, or build applications for any specific domain or sector. You can deploy them easily, and on top of that, you can create agents,” explained Vardhan.
“You can use your APIs effectively, integrating them in detail to build applications. This is where we believe we offer a complete solution, encompassing everything from the compute layer to the development of new applications,” he added.
Inside the AI studio, developers can use other LLMs also like Google’s Gemma 2 and Meta’s Llama 3.1.
SML launched text to text model Hanooman earlier this year in May. The model is named after the Hindu deity Hanuman. “Hanuman is a great example of responsible power. Despite being the most-powerful entity, he never used his power for selfish needs,” said Vardhan. As SML prepares for Hanooman’s launch, Vardhan remains optimistic about its broader implications.
