Red Hat is one of the world’s world giants Source and, as such, an authorized voice when pointing out how the open source AI should be, a model that has emerged strongly after the arrival of Deepseeek and that can be imposed in front to the closed, larger, but more efficient models.
Chris Wright, Chief Technology Officer and Senior Vice President of Global Engineering of Red Hat, explained the company’s vision on the AI open source in a very interesting post blog where he reviews everything related to the current thing of star technology and its Bet on smaller, optimized and open models of which they can be customized for cases for the use of specific business data anywhere and everywhere in the hybrid cloud.
The Executive recalls the importance of the Open Source in world technology and remembers that for three decades Red Hat saw the potential of how open source development and licenses could create a better software to boost innovation in you. That commitment to open source principles persists today, both in their business model and in corporate culture and believe that these concepts can have the same impact on artificial intelligence (AI) if they are applied in the right way.
However, there is a clear division within the world of technology in what is “the correct way.” AI, especially the great linguistic models (LLM) that drive generative AI, It cannot be seen in the same way that open source software. Unlike software, the AI models consist mainly in pesos of the model, which are numerical parameters that determine how a model processes the inputs, as well as the connections established between several data points. The pesos of the trained model are the result of an extensive training process that involves large amounts of training data that are prepared, mixed carefully.
While the pesos of the model are not software, in some aspects they fulfill a function similar to the code. It is easy to compare that the data is, or are analogous to the source code of the model. In the open source, the source code is commonly defined as the “preferred form” to make modifications to the software. Training data alone does not fit in this role due to their enormous volume and the complex pre -entry process. This process creates a dim and indirect connection between each training data, the trained weights and the final behavior of the model.
Most of the improvements and improvements of the AI models that are currently being carried out in the community They do not imply access or manipulation of original training data. Rather, they are the result of modifications in the pesos of the model or a fine adjustment process that can also be used to adjust the performance of the model. The freedom to make these improvements in the model requires that the weights be published with all the permits that users receive under open source licenses.
Red Hat’s vision on open source AI
Red Hat considers that The minimum threshold for open source AI is the pesos of the open source license model combined with open source software components. This is a starting point for open source AI, not the final destination. We encourage the open source community, to the regulatory authorities and the industry to continue striving to achieve greater transparency and alignment with the principles of open source development when training and adjusting the AI models.
This is Red Hat’s vision on how we, as an open source software ecosystem, can participate in a practical way in the AI of open source. It is not an attempt of a formal definition, such as the one carried out by the Open Source Initiative (OSI) with its definition of open source AI (Osaid). Our point of view to date is simply our opinion on what makes open source AI attainable and accessible to the broader complex of communities, organizations and suppliers.
We put this vision into practice through our work in open source communities, such as the Instructlab project, led by Red Hat, and our work around Granite family with open source license With IBM Research. Instructlab significantly reduces barriers to the contribution of AI models by professionals who are not dedicated to data science. With Instructlab, experts from various sectors can contribute their knowledge and skills, both for internal use and to boost an open and accessible open source AI model for UPSTREAM communities.
Chris Wright, Chief Technology Officer y Senior Vice President of Global Engineering de Red Hat
The family of Granite 3.0 models addresses a wide range of cases of use of AI, from the generation of code to the processing of natural language and the extraction of information from large data sets, all under a permissive open source license. We help IBM Research to bring a family of granite code models to the world of open source and continue to support the family of models, both from the point of view of the open source and part of our Red Hat AI offer.
The repercussions caused by the Recent Deepseek ads show how open source innovation can impact AIboth at the model level and in other aspects. Obviously, there are concerns about Deepseek’s approach, especially the lack of clarity in its license regarding the creation of the model, which reinforces the need for transparency. That said, this disruption validates our vision of the future of AI: one open, focused on smaller, optimized and open models that can be customized for cases for the use of specific business data anywhere and everywhere in the hybrid cloud.
Expanding open source AI beyond models
The technology and the principles of open source development are fundamental for our AI offers, as for our Red Hat AI portfolio. Red Hat Onshift AI is based on Kubernettes, Kubeflow, containers compatible with Open Container Initiative (OCI) along with a series of other native open cloud open source technologies. Red Hat Enterprise Linux AI (RHEL AI) incorporates the Granite LLM family of IBM, under open source license, and the Instructlab open source project.
The work of Red Hat in the field of open source AI goes far beyond Instructlab and the family of Granite models, covering the necessary tools and platforms for the consumption and productive use of AI. We actively participate in an increasing number of UPSTREAM projects and communities, and we have started many more on our own, including (among others):
- Ramalama, an open source project that seeks to simplify and facilitate the management and local IA models service.
- Trustyai, a set of open source tools to build more responsible workflows.
- Climatik, a project focused on helping AI to be more sustainable in terms of energy consumption.
- Podman AI LAB, a set of developers focused on facilitating experimentation with open source LLMs.
The recent announcement about Neural Magic reinforces our vision of AI, making it possible for organizations Align smaller and optimized AI modelsincluding open source license models, with their data, wherever they reside in the hybrid cloud. IT organizations can use the VLLM inference server to boost the decisions and results of these models, which helps build an AI stack based on transparent and compatible technologies.
Conclusions
For Red Hat, Open source AI lives and breathes in the hybrid cloud. The hybrid cloud provides the flexibility that is needed so much to choose the best environment for each AI workload, optimizing the performance, cost, scalability and safety requirements. Our platforms, objectives and organization support this effort, and we hope to collaborate with industry partners, our clients and the open source community in general as we continue to promote the innovation of open source AI.
There is a immense potential to expand this open collaboration in the AI space. We see a future that covers transparent work in the models, as well as their training. Either next week or next month (or before, given the speed with which AI evolves), we will continue to support and adopt the efforts that extend the limits of what it means democratize and open the world of AI.
