Recently, OS2ATC, an annual technology event in the field of open source operating systems, was held in Beijing, where many industry technologists and experts gathered to share the latest technical achievements and innovative ideas in the fields of Al and System, Hardware, Kernel, RISC-V Architecture, ARM Architecture, Longxin Architecture, Programming Technology, RUST, Intelligent Vehicles and Robotics, AIOT, Cloud Native, Virtualisation, and so on, focusing on the topic of "Open Source Intelligence".

Open Source Operating System Annual Technical Conference (OS2ATC) has been held for ten consecutive years, which has been effective in promoting the development of OS-related teaching, research and industry in China, strengthening academic exchanges, demonstrating industrial achievements, and launching interdisciplinary exchanges.

At the meeting, scholars and experts from Tsinghua University, Peking University, the National Innovation Centre for Intelligent and Connected Vehicles and related enterprises with outstanding achievements in the field of operating systems and other basic software brought year-end technical reports.

deepin was invited to participate in the keynote speeches of "AI and System" and "Programming Languages and System Ecology". Among them, UOS AI R&D Manager Zheng Youge will share "The Impact of AI Era on Operating System Interaction", and Open Source Community R&D Engineer Yang Chang will share "deepin's RISC-V / Loongarch Support: What are we talking about when we talk about desktop ecology?"


Operating System in the Age of AI

UOS AI R&D Manager Zheng Youge introduced the AI architecture, multimedia and retrieval AI technologies in deepin, explained the functional scenarios of AI landing in deepin, and deepin's exploration of the next-generation operating system and the outlook of AI convergence approach.

Since deepin announced in September 2023 that it had accessed large models to achieve intelligent upgrading of a number of self-study applications, it has continued to explore OS+AI, and has now achieved access to the capabilities of mainstream large models at home and abroad, and encapsulated them into deepin's AI bottom-up public capabilities to create AI assistants, which have been deeply integrated with mailboxes, browsers, global searches, graphs, and applications such as deepin-IDE, and have brought an intelligent experience to users.

AI in deepin is divided into four layers:

  • The hardware chip, end-side model, and end-side acceleration framework constitute the model layer, which mainly provides Low API capabilities;
  • The model management service, AI Agent service and RAG service module constitute the service encapsulation layer of High API;
  • The DTK API encapsulates the High API interface to facilitate application access. Currently, it mainly supports streaming local RPC encapsulation;
  • UOS AI serves as a user interaction portal, and each application as a plug-in is called by the Function Call capability of UOS AI, meanwhile, each application can also call the capability provided by UOS AI through the DBus interface.

Its features include:

1, application empowerment, improve user office efficiency

In daily office scenarios, manual and repetitive operations are the main reason why office efficiency is not ideal. UOS AI support is embedded in the application to quickly achieve application AI, and the call between applications is independent of each other, data isolation, no interference. Currently, the system's browser, mailbox, global search has been completed access, so that the user office to achieve "end-to-end" experience. In the future, it will be opened to ecological partners to help them realise AI capabilities out-of-the-box and reduce costs and increase efficiency.

2, Intelligent coding to improve the development efficiency of developers and eco-users

When developing AI applications, the development of AI-related public capabilities will take developers a lot of manpower and time. With the AI programming capability embedded in deepin-IDE, developers can use code generation, annotation generation, code translation, and code completion functions to make supplementary designs for their own application code, thus achieving twice the development efficiency with half the effort. Eco-partners can also use this capability to quickly build their own deepin applications, realising "zero-barrier" development.

3. Unified access to multiple models reduces development costs for ecological partners

When developing AI applications, if developers want to support multiple AI models, they need to dock them individually, which increases the overall development cost. The developer application access UOS AI, just through the UOS AI unified interface, docking multiple models, simplify the multi-model docking costs.UOS AI almost the current mainstream models have been adapted to access, such as Baidu, Xunfei, Smart Spectrum, 360 Smart Brain, ChatGPT large models, and so on.

4, custom prompt word, create your own AI function

Custom cue word function, support users through the cue word, has achieved personalised features, define different roles to meet the different scenarios of different users, to truly achieve "personalised" features.

In the future, deepin will continue to promote the improvement of AI capabilities, enhance interactivity, expand richness, improve openness, and lay a solid foundation for building the next-generation operating system.


deepin's RISC-V/LoongArch Support

As a builder of desktop operating system ecosystem, deepin has formed a close partnership with various chip, machine, middleware, database and other vendors, and has established cooperation with a wider range of eco-partners in terms of products and technologies, actively promoting various types of architecture adaptations, and contributing to the construction of a new operating system ecosystem in China.

Yang Chang, R&D Engineer of deepin open source community, shared the current status and challenges of RISC-V/LoongArch architectures in the Linux desktop ecosystem and deepin's efforts to address these issues at OS2ATC.

Starting in 2023, the deepin-ports SIG is continuously providing deepin V23 mainline-enabled adaptations for both RISC-V and LoongArch architectures, based on existing deepin-ports branch repositories, with mainline consolidation goals of:

  • Build out the basic toolchain for RV64/LA64 on the deepin mainline;
  • Fully merge patches from the deepin-ports repositories into the mainline;
  • Fully upgrade deepin-ports with the necessary high-version packages in the mainline;
  • The small portion of packages from deepin-ports that cannot be merged into the mainline are well-maintained in a branch of the mainline.

The deepin-ports SIG is committed to building a more unified, complete, and secure desktop ecosystem on multiple architectures. In the face of the current RISC-V/LoongArch ecosystem, which is divided by instruction set expansion, embedded ecosystem is not suitable for desktop, and the commercialised old world ecosystem is separated from the community-based new world ecosystem, deepin-ports tries to solve the ecological problems from the perspective of open-source community through various ways, continuous promotion of software package version upgrade, and support of the mainline software package architecture. The deepin-ports SIG will continue to provide strong support to ensure the full architecture of deepin V23.

The conference not only provided participants with a wealth of technical knowledge, but also pointed out the direction for the future development of open source operating systems. Through in-depth discussions and exchanges among experts, the conference has made significant contributions to the teaching, research and industrial development of China's open source operating systems and related fields. deepin is willing to work with you to build the Linux ecosystem and promote the continuous evolution of domestic operating systems!


Content source: deepin community
Reprinted with attribution

Leave a Reply