Huawei chips used to refine DeepSeek model in Chinese AI milestone
Researchers in China have achieved a significant milestone by utilizing Huawei processors for comprehensive post-training on DeepSeek's AI model, which boasts 1.6 trillion parameters. This process was executed on a computing cluster equipped with over 1,000 Huawei Ascend 910 chips, as reported by the Shenzhen government. The initiative involved collaborations among Huawei, the Shenzhen Loop Area Institute, Harbin Institute of Technology's Shenzhen campus, and the Shenzhen Research Institute of Big Data. This advancement enhances local computing power and represents a step towards China's goal of greater self-reliance in artificial intelligence technology.
Key facts
- DeepSeek's largest model has 1.6 trillion parameters.
- At least 1,000 Huawei Ascend 910 chips were used.
- Full-parameter post-training was successfully conducted.
- The project was announced by the Shenzhen government on Friday.
- Collaborators include Huawei, Shenzhen Loop Area Institute, Harbin Institute of Technology (Shenzhen campus), and Shenzhen Research Institute of Big Data.
- Chinese chipmakers have struggled with AI training but succeeded with inference.
- Post-training refines a model to follow instructions and safety rules.
- The achievement aims to boost China's AI self-reliance.
Entities
Institutions
- Huawei
- DeepSeek
- Shenzhen Loop Area Institute
- Harbin Institute of Technology
- Shenzhen Research Institute of Big Data
- Shenzhen government
Locations
- China
- Shenzhen