PyTorch + Shengteng jointly promote the innovative development of AI ecosystem

On October 5, 2023 (Beijing time), the PyTorch community officially released version 2.1. Through the continuous cooperation and joint efforts between the PyTorch community and Ascend, PyTorch version 2.1 has simultaneously supported Ascend NPU. This means that developers can develop Ascend-based models directly on PyTorch 2.1. In the future, through continued community technology contributions, Ascend will evolve and release simultaneously with the PyTorch community version, jointly promoting AI technology innovation and open source ecological development

Illustration: The official release statement of PyTorch community version 2.1 announced that Ascend NPU has been successfully connected

PyTorch has released a more complete third-party device access mechanism in the new version. This feature is led by Ascend and completed together with the core maintainers of the PyTorch community. Based on this feature, third-party AI computing devices can connect to the PyTorch framework without modifying the original framework code. Ascend also provides an officially certified reference implementation of Torch NPU, which can guide third-party devices to easily access it.

In earlier versions, due to the lack of native support for Ascend computing devices, developers would need to spend several weeks migrating models if they wanted to use Ascend NPU to achieve acceleration under mainstream frameworks. Now, based on the new version, users can directly enjoy the native PyTorch development experience on Ascend NPU, and obtain models and applications that can run efficiently on Ascend computing devices.

Shengteng has adapted BLOOM, GPT-3, LLaMA and other mainstream large models in the industry on PyTorch, and has deeply optimized them to ensure that the performance is on par with the industry. On the latest PyTorch version, Ascend will work closely with the community to evolve simultaneously and continuously improve model acceleration capabilities

In the future, Ascend will actively promote the improvement of the diversity of computing power in the PyTorch community through continuous technical contributions, optimize the PyTorch framework, help release the computing power of Ascend, and work with PyTorch community contributors to promote AI technology innovation and the rapid development of the open source ecosystem develop.

The above is the detailed content of PyTorch + Shengteng jointly promote the innovative development of AI ecosystem. For more information, please follow other related articles on the PHP Chinese website!