More than three decades ago, Red Hat saw the potential of open-source development and licenses to create better software and foster IT innovation. Thirty million lines of code later, Linux has not only evolved to become the most successful open-source software but also maintains that position today. The commitment to open-source principles continues, not only in the corporate business model but is also part of the work culture. In the company’s assessment, these concepts have the same impact on artificial intelligence (AI) if done correctly, but the technology world is divided on what would be the ‘right way’.
AI, especially large language models (LLMs) behind generative AI, cannot be viewed in the same way as open-source software. Unlike software, AI models consist mainly of numerical parameter models that determine how a model processes inputs, as well as the connections it makes between various data points. Trained model parameters are the result of a lengthy process involving vast amounts of training data that are carefully prepared, mixed, and processed.
Although model parameters are not software, they do have a function similar to code in some respects. It is easy to liken data to be the model’s source code, or very close to it. In open source, the source code is commonly defined as the ‘preferred form’ for making modifications to the software. The training data alone does not fit into this function, given their different size and the complicated pre-training process that results in a tenuous and indirect connection that any item of the data used in training has to the trained parameters and resulting model behavior.
Most of the improvements and enhancements in AI models happening now in the community do not involve accessing or manipulating the original training data. Instead, they result from modifications to model parameters or a process or tuning that can also serve to adjust the model’s performance. The freedom to make these model improvements requires that parameters be released with all the permissions users receive under open-source licenses.
Red Hat’s Vision for Open Source AI.
Red Hat believes that the foundation of open-source AI lies in model parameters licensed open-source combined with open-source software components. This is a starting point for open-source AI, but not the ultimate destination of the philosophy. Red Hat encourages the open-source community, regulatory authorities, and the industry to continue striving for greater transparency and alignment with open-source development principles when training and fine-tuning AI models.
This is Red Hat’s view as a company that encompasses an open-source software ecosystem, can practically engage with open-source AI. It is not an attempt at a formal definition, like the one the Open Source Initiative (OSI) is developing with its Open Source AI Definition (OSAID). This is the corporate perspective on how open-source AI makes it feasible and accessible to the widest range of communities, organizations, and providers.
This perspective in practice is put into practice through work with open-source communities, highlighted by the InstructLab project, led by Red Hat and the effort with IBM Research in the Granite family of licensed open-source models. InstructLab significantly reduces the barriers for non-data scientists to contribute to AI models. With InstructLab, domain experts from all sectors can add their skills and knowledge, both for internal use and to help a shared open-source AI model widely accessible to upstream communities.
The Granite 3.0 model family deals with a wide range of AI use cases, from code generation to natural language processing to extract insights from large datasets, all under a permissive open-source license. We helped IBM Research bring the Granite code model family to the open-source world and continue to support the model family, both from an open-source perspective and as part of our Red Hat AI offering.
The repercussion of the recent DeepSeek ads shows how open-source innovation can impact AI, both at the model level and beyond. There are obviously concerns about the Chinese platform’s approach, mainly that the model’s license does not explain how it was produced, which reinforces the need for transparency. That being said, the disruption mentioned reinforces Red Hat’s vision of the future of AI: an open future, focused on smaller, optimized, and open models that can be customized for specific enterprise data use cases anywhere in the hybrid cloud.
Expanding AI models beyond open source
Red Hat’s work in the open-source AI space goes far beyond InstructLab and the Granite model family, extending to the tools and platforms needed to actually consume and productively use AI. The company has become very active in fostering technology projects and communities, such as (but not limited to):
● RamaLama, a open-source project aimed at making it easier to manage and deploy AI model locations;
● TrustyAI, an open-source toolkit for building more responsible AI workflows;
● Climatik, a project focused on helping make AI more sustainable when it comes to energy consumption;
● Podman AI Lab, a developer toolkit focused on simplifying experimentation with open-source LLMs;
The recent announcement about Neural Magic expands the corporate view on AI, making it possible for organizations to align smaller, optimized AI models, including licensed open-source systems, with their data, wherever it resides in the hybrid cloud. IT organizations can then use the inference server vLLM to drive decisions and the production of these models, helping to build an AI stack based on transparent and supported technologies.
For the enterprise, open source AI lives and breathes in the hybrid cloud. The hybrid cloud provides the necessary flexibility to choose the best environment for each AI workload, optimizing performance, cost, scale, and security requirements. Red Hat’s platforms, goals, and organization support these efforts, along with industry partners, customers, and the open source community, as open source in artificial intelligence continues to be propelled.
There is immense potential to expand this open collaboration in the AI space. Red Hat sees a future that encompasses transparent work on models, as well as their training. Whether it’s next week or next month (or even sooner, given the speed of AI evolution), the company and the open community as a whole will continue to support and embrace efforts to democratize and open up the world of AI.
