As we all know, the application of deep learning open source framework is pushing artificial intelligence technology from the laboratory to the industry. As an important production tool in the AI era, the AI open source framework is an important cornerstone for developers to develop AI applications, and is also a companion on the development path of developers.
However, looking at the open source frameworks on the market, TensorFlow and PyTorch are basically divided into two parts, from technical capabilities to ecological construction, which can satisfy most enterprises to build their own AI applications. On the other hand, the domestic AI open source framework is still difficult to compete with the first two in terms of market influence and scale of use.
Even so, the development of the AI framework in China has become increasingly lively.
In 2016, Baidu PaddlePaddle fired the first shot of open source domestic AI framework. In 2020, AI open source frameworks from China will be on the stage one after another. Tsinghua, Megvii, Huawei, the four first-class science and technology universities and industry organizations have successively announced open source AI frameworks MindSpore, MegEngine, Jittor, OneFlow, which may become domestic deep learning. An important highlight in the history of framework open source.
Undoubtedly, these domestic AI open source frameworks are all referring to TensorFlow and PyTorch. Before the market structure of AI open source frameworks is completely solidified, they will all have the opportunity to take a share of these overlords, and even form a three-sided competition with TensorFlow and PyTorch. situation.
So, in these AI open source frameworks, who will have the opportunity to compete with TensorFlow and PyTorch?
At present, based on their respective technical advantages, each open source framework has begun to take shape, and continue to expand from the product and ecology, step by step. Among them, Megvii Tianyuan MegEngine has continued to make great strides in technology and ecology after open source.
Tianyuan is a framework that accompanies Megvii's own actual combat experience in the AI industry and is one of the core components of Megvii Brain++. It began research and development in 2014 and was used by all employees in 2015. When the open source was provided to global developers in March this year, Megvii made a comprehensive upgrade for Tianyuan.
Not only can it be upgraded and buffed for despise and monsters in the AI competition arena, it also supports half of the sky of despise engineering and productization. At present, all Megvii algorithms are trained and inferred through Tianyuan MegEngine.
In industrial practice, they continued to iterate on the underlying framework, data and data facilities, and finally completed the full migration from R&D to business to their own deep learning framework and own computing cluster. In the past few years, Megvii has encountered many common pain points in the industry during the research and development process, and Tianyuan’s core feature is to solve these pain points.
Subsequently, Tianyuan MegEngine continued to change, from the Alpha version in March to the Beta version in June, to the 1.0 preview version in September, and it went through 8 iterations. After half a year of technical iteration from open source to now, Tianyuan has three core advantages: integrated training and reasoning, efficient support on the entire platform, and combined training capabilities.
In use, Tianyuan Model Hub provides a wealth of pre-trained models, and each model provides SOTA-level accuracy, so that developers can easily get started with Tianyuan. In addition, Tianyuan has been deeply integrated with Xiaomi MACE and OpenAI Lab Tengine. Developers can directly convert MegEngine models to MACE or Tengine for execution, thereby gaining the ability to execute deep learning models on various heterogeneous devices.
GitHub address:
https://github.com/MegEngine
With the continuous development of Tianyuan MegEngine, it has gradually become well-known by developers and has helped Chinese companies and developers to implement AI applications. Now it has 3400 Stars from developers on GitHub.
At present, Tianyuan MegEngine is still evolving towards the goal of the star of China's AI open source framework. In the future, who do you think will lead the domestic AI open source framework, and what opportunities does Tianyuan MegEngine have?
The current development of Tianyuan MegEngine is inseparable from the support of developers. In order to thank the developers for their assistance to the Chinese original open source framework, this time, we will give back to the developers through the activity "Support domestic AI open source frameworks, developers are in action!", all Star Tianyuan MegEngine users on GitHub have the opportunity to get cool Tianyuan MegEngine commemorative T-shirt, CSDN customized keyboard support, exquisite AI technical books and 100-hour online computing card, developers are welcome to actively participate in the like activity.
GitHub link:
https://github.com/MegEngine/MegEngine
Participate now:
https://bbs.csdn.net/topics/397736554
You can also scan the QR code to participate in the event
Attachment: GPU computing power conversion method:
Step1: Register as a MegStudio user (https://studio.brainpp.com), and provide MegStudio registered account to CSDN staff;
Step 2: We will send 100 hours of GPU computing power to the accounts of the award-winning users for free, and then we can start the deep learning model training journey
Participating partners in this event need to like and register as a MegStudio user, take screenshots and reply to the post to receive the following prizes.
If you have any questions, you can also send us an email, email: [email protected]