Huawei Open-Sources Pre-Training, SFT and RL Code for openPangu-2.0

Huawei has announced that the pre-training, SFT code, and post-training RL code for its open-source Pangu openPangu-2.0 are now officially open-sourced. OpenPangu is Huawei's open-source AI model brand, aiming to provide best practice references for using Ascend through native training and inference technology.

The openPangu-2.0 model components have been gradually released on open-source platforms, with code hosted on GitCode's Ascend Tribe repository. This open-source release covers the full pipeline from pre-training to post-training reinforcement learning, allowing developers to train and run inference on the Ascend platform.

This move provides AI developers with a complete training reference based on the Ascend ecosystem, helping lower the barrier to large model development and promoting community co-building and application of Ascend-native AI models.