IBM announced the release of its latest suite of artificial intelligence (AI) models, Granite 3.0, during its annual TechXchange event in Las Vegas. The third-generation models, part of the Granite flagship series, are designed to perform on par with similar-sized models from other providers, said IBM.
According to Big Blue, these models have been benchmarked against various academic and industry standards, focusing on performance, transparency, and safety. The Granite models are available under the Apache 2.0 licence, aligning with IBM’s open-source AI approach, which offers enterprise users and the community the flexibility and autonomy to implement the models as needed.
The Granite 3.0 series includes general-purpose language models, safety-focused models, and mixture-of-experts models. These models are built for tasks such as retrieval-augmented generation (RAG), classification, summarisation, and entity extraction, and can be fine-tuned with enterprise-specific data for integration into various business workflows.
IBM’s latest AI models also feature built-in safety mechanisms, providing risk assessments of user prompts and responses. Additionally, the models are equipped to identify risks such as social bias and harmful content. They also perform checks tailored to RAG use cases, ensuring information relevance and contextual accuracy.
Big Blue has also introduced Mixture-of-Experts models, designed for deployment in low-latency environments, making them suitable for CPU-based and edge setups.
These models support enterprise data and incorporate smaller Granite models using the alignment technique InstructLab, developed in partnership with Red Hat. IBM claims that early proofs of concept demonstrate cost reductions ranging from three to 23 times compared to larger models.
The Granite 3.0 models are accessible through platforms like HuggingFace, IBM’s watsonx, Nvidia NIM microservices, and additional cloud services such as Google Cloud’s Vertex AI Model Garden, Ollama, and Replicate.
Furthermore, IBM’s strategic partnerships with AWS, Salesforce, and SAP provide enhanced integration options for enterprise users.
IBM’s AI lag draws industry criticism
Despite IBM’s advancements in AI technology, including platforms like watsonx, the company has faced criticism for lagging behind competitors such as Microsoft and Google in terms of effectiveness and monetisation. Its plans to replace 7,800 roles with AI were also criticised by sources within the company as unrealistic.
IBM CEO Arvind Krishna acknowledged in a December 2023 interview with CNBC that the company had been slow to monetise its AI advancements. Krishna explained that the company’s approach involved developing large, complex solutions, which he admitted were not suitable for market adoption at the time.
Read more: IBM launches AI foundation model for weather and climate applications
