DeepSeek Open-Sources Infrastructure for Huawei Ascend Computing Platform

DeepSeek, an artificial intelligence company, has announced the open-sourcing of key infrastructure components for the Huawei Ascend computing platform. This release includes the TileLang high-level language compilation tool, various computing libraries, and distributed communication libraries. These components are designed to enhance development efficiency and simplify programming logic while maintaining high hardware performance, similar to previously released components for the NVIDIA platform.

TileLang is intended to be a general-purpose programming language that can fully utilize hardware capabilities. It has already been validated on the NVIDIA platform, supporting the training of DeepSeek's V4 series models. The newly open-sourced Ascend version encapsulates the underlying instructions of the Ascend C, allowing developers to use a high-level programming approach without compromising on performance. Each TileLang operator utilized in DeepSeek's training has a corresponding high-performance implementation on the Ascend platform.

In addition to TileLang, the open-sourced components include DeepGEMM, DeepEP, TileKernels, FlashMLA, and DeepSelect. These libraries facilitate various functions such as general matrix operations, large-scale cross-device communication, conventional vector calculations, long-context sparse attention, and data filtering. DeepSeek reports that performance in key computing and communication tests has approached hardware limits, thanks in part to collaboration with Huawei, which supported the development of a 128-card supernode solution based on the Ascend 950.

© 2026 KLEA News. All Rights Reserved. This article is provided for informational purposes only. It is not offered or intended to be used as legal, tax, investment, financial, or other advice.

Source: KLEA News