Small language models and AI agents are the talk of the AI town. With the release of the latest Granite 3.0 and Bee Agent Framework, IBM has shifted its focus towards smaller and more efficient models while maintaining transparency to a level previously unheard of in the AI industry.
A few days after the launch of Granite 3.0, Armand Ruiz, VP of product–AI Platform at IBM, publicly disclosed the datasets on which the model was trained. “This is true transparency. No other LLM provider shares such detailed information about its training datasets,” said Ruiz.
This is a practice IBM has adhered to even in the past with new model releases. Vishal Chahal, VP of IBM India Software Lab, told AIM that it is important for companies to be fully open about their language models and the data they’ve used to build them.
Despite its smaller size, Granite 3.0 has been performing exceptionally well in tasks like coding and making small chatbots. The model is applicable across diverse domains, from healthcare to finance, demonstrating its adaptability. Looks like open-source models have finally caught up with their closed-source counterparts.
“You can start small,” Chahal highlighted, sharing that Granite 3.0 can even run on a Mac. This flexibility allows businesses to scale as their needs grow without initially needing extensive GPU clusters.
The open-source philosophy extends beyond Granite’s datasets to its availability on platforms like GitHub, Hugging Face, and IBM WatsonX. IBM’s AI alliance with partners such as Meta, LlamaIndex and Ollama further broadens accessibility for developers worldwide.
Open Source–The Backbone of Granite’s Success in India
India’s developer ecosystem has embraced Granite 3.0 wholeheartedly. Thousands of developers are already leveraging it, drawn by its performance and versatility. Beyond individual developers, significant collaborations are happening at the institutional level. “India’s developer ecosystem is vibrant and growing,” Chahal said, “and we are committed to supporting it every step of the way.”
“Institutions like IIT Madras, IIT Bombay, and IIT Jodhpur have become part of IBM’s AI alliance,” Chahal shared. These institutions, along with initiatives like AI4Bharat, are using Granite to create benchmarks for Indian languages like MILU, which was released last month.
Major players like Infosys, KissanAI, Wadhwani AI, and Sarvam AI have also integrated Granite into their AI ecosystems, demonstrating its relevance to startups and enterprises alike. This collaboration is fostering innovation tailored to India’s unique needs, from agriculture to vernacular language processing.
Chahal explained that many startups are using Granite alongside other models, such as Meta’s Llama. “Granite is exceptional for tasks like coding instructions,” he said, “but a multi-model approach ensures startups can pick the best model for each use case.” IBM’s WatsonX platform facilitates this by hosting both IBM’s and partner models, allowing developers to choose the most suitable tools for their needs.
Chahal said that IBM plans to release a multimodal model soon, which will include speech and video models.
Beyond just language models, at its recent TechXchange conference, IBM unveiled advancements in AI agentic capabilities through frameworks like the Bee Agent and MARC (Multi-Agent Resource Coordination) frameworks. “We are officially in the agentic world,” Chahal declared.
IBM offers multiple ways to build agents, from WatsonX and Watson Orchestrate to WatsonX Flows, catering to diverse business requirements. “Soon, we will see a multi-agent world, where hundreds of agents work collaboratively across enterprises,” he predicted.
Meanwhile, IBM’s Bee Agent Framework is designed to assist developers in creating powerful agents with minimal adjustments to existing implementations, particularly as it actively works to optimise for other popular LLMs.
The framework includes key features for building versatile agents. Its built-in Bee agent, configured for Llama 3.1, is ready to use, though users can also customise their own agents using built-in tools in JavaScript or Python.
Supporting the Developers
“For developers, trust is paramount. A model should transition seamlessly across the development, testing, and production environments without requiring constant adjustments,” said Chahal, adding that the ability to retrain models efficiently is also critical for developers.
Retraining should not require duplicating models and consuming excessive resources.
At IBM Red Hat’s InstructLab, the team has developed an open-source tool that allows developers to retrain models using a single GPU cluster. By adding knowledge layers tailored to industry-specific needs, this tool eliminates the need for multiple copies and streamlines the retraining process. For developers, this solves a significant challenge: managing infrastructure while scaling capabilities.
“We aim to lower the barriers for developers by providing access to smaller, efficient models that work even on a single GPU or CPU cluster,” said Chahal, while adding that beyond AI models, developers also need robust data tools and pipelines. This is where IBM’s research software and systems labs collaborate with colleges and universities to share these tools and empower developers with end-to-end capabilities.
In July, IBM held an International GenAI Conclave in Kochi, where college students competed in a hackathon, and winners were recognised by dignitaries. IBM also hired some of these developers, demonstrating its commitment to nurturing talent.
The company also launched a GenAI Innovation Centre in Kochi, which allows colleges, universities, startups, and partners to experiment with business use cases and learn how GenAI is applied in real-world scenarios. This centre will soon be replicated in other locations, creating open zones for innovation.
“Building in India, for India, Bharat, and the world,” Chahal said, adding that IBM wants Indian developers to aspire to have a global impact. With IBM’s support, they can create solutions that not only address local challenges but also have global relevance.
The post Granite 3.0 is IBM’s Love Letter to Indian Developers appeared first on Analytics India Magazine.
