The time has come, that moment in which it is worthwhile to take a look around and observe not only the subject under discussion but also the conceptual bases from which it is being built, evolved and iterated: we are referring to the two major hegemonic software development paradigms Open Source and Closed Source and their interrelation in the context of Artificial Intelligence.
If we look one step below the big corporations or projects related to AI and ask ourselves some related questions about its technical and, therefore, popular emergence, we will inevitably arrive at the software layer that allows interacting with the models. Currently there are two major frameworks, PyTorch (Meta) and TensorFlow (Google) each with its own characteristics, differences and two aspects that are identical:
- Both were created by large technology companies.
- Both opted for the Open Source path and their communities.
The PyTorch case
In 2016, Facebook (now Meta), launched PyTorch as an open source project to facilitate the development of machine learning models. The goal was to create a platform that would allow researchers and developers to work together on creating more efficient algorithms and models in a more efficient way.In its early days, PyTorch quickly became a popular project among the machine learning community. The platform offered a simple, easy-to-use syntax, allowing developers to create and interact with complex models with ease.
In 2017, Nvidia joined the PyTorch project as one of its main sponsors. With its expertise in GPU hardware, Nvidia helped to significantly improve the performance of the framework and make it more compatible with its own products.
In 2022, the PyTorch Foundation was established as a non-profit organization encompassing several large companies, including Nvidia, AMD, Microsoft, AWS and Meta. The foundation aims to promote the development and adoption of PyTorch worldwide.
The Open Source community
The community around PyTorch has been fundamental to its success. Developers have created a large number of libraries, tools and models that can be used with the framework, as well as contributed to the creation of documentation, tutorials and educational resources, online help. This has allowed its constant evolution and adaptation to the needs of the field in which it operates.
Looking at the history of the interrelationship between Enterprise(s), Academia and the Open Source community leads to the conclusion that this relationship is viable and provides solutions with end products that ultimately move the industry forward. This being true, on a second glance we should not forget the motivations of each of the sectors in this collaboration. We do not intend to assess or judge whether they are good or bad, we simply intend that they should not be omitted.
Standards and motivations
There are several reasons why a large corporation might decide to create an open source project like PyTorch:
- Access to the community: By launching an open source project, the corporation can access a community of developers and users committed to the project, which can lead to faster and more efficient improvements.
- Reduce costs: Software development is an expensive process, but by utilizing community resources and collaborating with other developers, the corporation can reduce its research and development expenses.
- Increase visibility: An open source project can increase the corporation’s visibility as a leader in its industry or specific field, which can lead to greater credibility and trust with customers and partners.
- Encourage innovation: Open collaboration and open source access can foster innovation and allow other developers to contribute new ideas and solutions, which can lead to significant product or service improvements.
- Access to additional resources: By launching an open source project, the corporation can access additional resources such as investment funds, human talent and strategic collaborations with other companies and organizations.
- Collaboration with other projects: An open source project can collaborate with other similar or related projects, which can lead to greater integration and compatibility between different solutions.
- Access to data: By launching an open source project, the corporation can access data and knowledge from the community, which will be useful to improve the product or service.
In the specific case of PyTorch, Meta created the project as a way to facilitate the development of machine learning models and allow other developers to contribute new ideas and solutions. By launching PyTorch as an open source project,
By launching PyTorch as an open source project, Facebook was able to:
- Establish itself as a leader: By creating a popular and widely used framework for developing machine learning models, Facebook established itself as a leader in the field of AI. TensorFlow is also very popular, although it should be noted that Google implemented Pytorch in its Cloud.
- Positioning itself as a key player (standards): By being the creator of the PyTorch project, Facebook positioned itself as a key player in the AI field, allowing it to influence the direction of the development of AI-related technologies.
The widespread adoption of PyTorch has also led to:
- Unification of standards: Projects and companies are adopting PyTorch as the standard for the development of machine learning models, which has unified the industry around a common platform.
- Continuous improvement: Open collaboration and the contribution of external developers have allowed PyTorch to be continuously improved, leading to significant improvements in its performance and functionality.
Actions by Meta, such as the release of the Llama 3.1 model (8B, 70B and 405B) under the Open Source premise, indicate that this is the path they are going to follow.
At this point it should be noted that, despite its impetuous collaboration, both with other companies in the sector and with Open Source, Meta is a private company, one of the world’s technological giants, and this marks a line that simply needs to be remembered from time to time. On the other hand, there is the well-known doubt between fragmentation and centralization, of key parts such as the case in question.
Official PyTorch Documentary: Powering the AI Revolution
Not Just PyTorch and TensorFlow: 4 Other Deep Learning Libraries You Should Know | towardsdatascience.com
