19 hours Ago By Franklin Okeke
IBM has unveiled its newest suite of generative AI models, “Granite 3.0,” aimed at enterprise clients. Granite 3.
0 models, like predecessors in the Granite family, will be open-source under Apache 2.0 license. In a press release , IBM said the Granite 3.
0 large language model (LLM) can outperform or match similarly sized models from leading model providers on various academic and industry benchmarks. Alongside the free models, IBM offers Watsonx, a paid platform that allows organizations to run and personalize some variants of the Granite 3.0 AI models in their own environments.
Dario Gil, IBM’s Director of Research, said the Granite 3.0 models were developed using Nvidia’s advanced H100 GPUs. Granite 3.
0 Performance Benchmarks IBM claims its Granite 3.0 models were trained on over 12 trillion tokens across multiple programming languages and utilize a two-stage training method optimized through thousands of experiments. The Granite 3.
0 8B Instruct model variant outperforms other open-source models of a similar size from companies like Meta and Mistral, as assessed by Hugging Face’s OpenLLM Leaderboard, IBM said. For enterprise-focused tasks such as retrieval-augmented generation (RAG), tool use, and cybersecurity, the Granite 3.0 8B Instruct model leads in average performance, the company said.
The entire suite of Granite 3.0 models and the updated time series models are available for download on HuggingFace, with additional variants expected to be released soon. In recent months, IBM has doubled its efforts in AI research.
It recently announced the release of its forecasting model, TinyTimeMixer (TTM) AI. In a recent interview with Techopedia , IBM said the TTM AI is a lightweight AI model that does not rely on a gigantic number of parameters for optimal performance in specific domains..
