Soket AI Labs, an innovative Indian AI research firm, has embarked on a strategic partnership with Google Cloud to enhance its groundbreaking multilingual foundation model, Pragna-1B. This collaboration is set to significantly advance the adoption of generative AI in India, with a particular focus on supporting vernacular languages such as Hindi, English, Bengali, and Gujarati. The partnership aims to empower organizations and developers with advanced AI capabilities, making sophisticated AI tools more accessible and efficient.

Enhancing Pragna-1B with Google Cloud

The integration with Google Cloud represents a pivotal milestone for Soket AI Labs. Abhishek Upperwal, the visionary Founder of Soket AI Labs, emphasized the transformative impact of Google Cloud’s advanced AI infrastructure on Pragna-1B. Leveraging Google Cloud’s powerful computational resources, Pragna-1B has achieved remarkable efficiency and performance, even though it is trained on fewer parameters compared to similar models. Upperwal highlighted, “Pragna-1B offers balanced language representation and enables faster and more efficient tokenization, making it ideal for organizations seeking optimized operations and enhanced functionality. This collaboration with Google Cloud not only enhances the model’s capabilities but also underscores our commitment to driving AI innovation in India.”

The Pragna-1B model, which was initially released on May 1, 2024, has been designed to facilitate the widespread adoption of generative AI across India. By supporting multiple vernacular languages, Pragna-1B aims to bridge the language gap and provide robust AI solutions tailored to the diverse linguistic landscape of India.

Making Pragna-1B More Accessible

A key objective of the collaboration is to make Pragna-1B more accessible to developers and organizations. Soket AI Labs plans to list its AI Developer Platform on the Google Cloud Marketplace, along with the Pragna series of models on the Google Vertex AI model registry. This strategic move aims to provide a seamless and integrated experience for developers, enabling them to fine-tune models using high-performance resources such as Vertex AI and TPUs. The intuitive interface of Soket’s AI Developer Platform, combined with the robust computational power of Google Cloud, ensures optimal efficiency and scalability for AI projects.

Upperwal elaborated on the benefits of this integration, stating, “By making our AI Developer Platform and Pragna series available on the Google Cloud Marketplace, we are empowering developers with the tools they need to innovate and create. The combination of our specialized AI models and Google Cloud’s high-performance resources provides a powerful foundation for developing and deploying advanced AI solutions.”

Focus on Indian Context and Vernacular Languages

Pragna-1B has been meticulously designed with the Indian context in mind, ensuring that it meets the unique needs and challenges of enterprises operating in India. The model features a Transformer Decoder-only architecture with 1.25 billion parameters and a context length of 2048 tokens. It has been trained on approximately 150 billion tokens, with a particular focus on Hindi, Bangla, and Gujarati. This tailored approach ensures that Pragna-1B delivers state-of-the-art performance for vernacular languages, making it a valuable tool for promoting AI innovation in India.

The emphasis on vernacular languages is a significant aspect of Pragna-1B’s design. By providing robust support for languages such as Hindi, Bengali, and Gujarati, Pragna-1B addresses the linguistic diversity of India and ensures that AI solutions are accessible to a broader audience. This focus on language representation and tokenization efficiency makes Pragna-1B particularly well-suited for organizations seeking to integrate AI into their operations.

Soket AI Labs’ Journey and Future Plans

Founded by Abhishek Upperwal in 2019, Soket AI Labs has rapidly evolved from a decentralized data exchange for smart cities to a leading AI research firm. One of the key milestones in their journey was the creation of the ‘Bhasha’ series of high-quality datasets designed for training Indian language models. This includes ‘Bhasha-wiki,’ which comprises 44.1 million articles translated from English Wikipedia into six Indian languages, and “Bhasha-wiki-indic,” a refined subset focusing on content relevant to India. These datasets have played a crucial role in advancing AI research and development in India.

In a recent LinkedIn post, Upperwal discussed the significant improvements in GPT-4o’s tokenizer and vocabulary size, which now supports 200k tokens. However, he pointed out that Pragna-1B’s tokenizer still outperforms GPT-4o for languages like Kannada, Gujarati, Tamil, and Urdu. This ongoing commitment to improving language support and tokenization efficiency motivates Soket AI Labs to continually enhance the capabilities of Pragna-1B and expand its support for Indian languages.

Looking ahead, Soket AI Labs is experimenting with a Mixture of Experts model, which aims to expand the range of languages supported and explore different architectural approaches for increased optimization. This continuous innovation underscores Soket AI Labs’ dedication to advancing AI technology and ensuring that it meets the evolving needs of users and organizations in India.

Technical Collaboration and Innovation

The collaboration between Soket AI Labs and Google Cloud also extends to technical work on training large-scale models and curating high-quality datasets for Indian languages. This joint effort aims to promote AI innovation in India while ensuring transparency and clarity in the development process. By leveraging Google Cloud’s advanced AI infrastructure, Soket AI Labs can achieve greater efficiency and cost-effectiveness in developing and deploying AI models.

Bikram Singh Bedi, Vice President and Country Managing Director of Google Cloud India, expressed his enthusiasm for the partnership, stating, “We are thrilled to partner with Soket AI Labs to democratize AI innovation in India. Built on Google Cloud, the launch of Pragna-1B marks a pioneering leap in Indian language technology, offering enhanced scalability and efficiency for organizations. This collaboration underscores our commitment to supporting AI development in India and providing the tools and resources needed to drive innovation.