Archive — CN — technode
Huawei open-sources openPangu-2.0 pretraining, SFT and RL code
Huawei's Ascend Tribe has released the open-source code for the openPangu-2.0 model family, enabling pretraining, supervised fine-tuning, and reinforcement learning. This code supports Huawei’s Ascend-based training ecosystem, enhancing accessibility and flexibility in AI model development.
Key points
- Huawei released open-source code for openPangu-2.0, a model focused on AI.
- The code includes capabilities for pretraining, supervised fine-tuning (SFT), and reinforcement learning (RL).
- openPangu-2.0-Training supports both pretraining and SFT.
- openPangu-2.0-RL emphasizes reinforcement learning post-training.
- The projects aim to enhance the Ascend-based training ecosystem of Huawei.
Why it matters
- Open-source AI initiatives have been growing, aligning with the 2025 trend of democratizing AI access.
- Huawei's move reflects the increasing focus on AI model interoperability and collaboration in tech.
- The release responds to the ongoing global shift toward open-source software in the tech industry.
Huawei's open-source release of the openPangu-2.0 model supports the trend toward collaborative and accessible AI development, potentially influencing future AI advancements.