In a significant move towards fostering greater collaboration, Huawei has revealed its strategy to open-source its complete AI software stack. This initiative aims to empower developers, tackle previous hurdles, and cultivate robust community engagement. The roadmap outlines specific timelines and technical details for making its AI software accessible to the public.
At the heart of this open-source push is CANN (Compute Architecture for Neural Networks), Huawei’s core AI toolkit. The company plans to provide open interfaces for its compiler and virtual instruction set, while fully open-sourcing other CANN software elements. Furthermore, the Mind series of application enablement kits and toolchains will also be made openly available.
Huawei has also pledged to open-source its openPangu foundation models. Specific capabilities and licensing terms associated with these models are still to be determined. To ensure flexibility and integration with various operating systems, Huawei will release the entire UB OS Component under an open-source license, facilitating its adoption within systems like openEuler.
Recognizing the importance of framework compatibility, Huawei is prioritizing support for popular open-source communities like PyTorch and vLLM. While a release is anticipated by December 31, the initial quality of the open-sourced software will heavily influence community reception. Crucially, decisions regarding license selection and overall governance structures are yet to be finalized. The upcoming months present a valuable opportunity for developers to assess and prepare for Huawei’s open-source AI development platform.