Ant Group Utilizes Local GPUs Over Nvidia to Reduce AI Costs
Cost-Effective AI Training with Local GPUs
Ant Group, the fintech arm of Alibaba Group, has made significant strides in training large language models (LLMs) by utilizing locally produced graphics processing units (GPUs). This groundbreaking approach has resulted in a 20% reduction in training costs, as stated in recent research.
Innovative Model Development
The team behind Ant's Ling-Plus-Base model, a Mixture-of-Experts (MoE) architecture with 300 billion parameters, has found that it can be trained on less powerful devices without compromising performance. Results show that by avoiding high-end Nvidia GPUs, computational expenses decreased significantly during pre-training, maintaining effectiveness similar to models like Qwen2.5-72B-Instruct and DeepSeek-V2.5-1210-Chat.
- Algorithm flexibility: This showcases the potential of using lower-performance hardware for substantial machine learning tasks.
- Industry positioning: Ant Group stands alongside peers like DeepSeek and ByteDance, mitigating reliance on Nvidia amidst stringent US export regulations.
- Adaptive strategies: Chinese tech companies are innovating around advanced labor-saving measures due to ongoing US-China tech tensions.
Healthcare AI Innovations
Additionally, Ant Group announced major upgrades to its AI solutions for the healthcare sector, enhancing their offerings to combine medical reasoning and multimodal interaction in comprehensive “all-in-one machines.”
This article was prepared using information from open sources in accordance with the principles of Ethical Policy. The editorial team is not responsible for absolute accuracy, as it relies on data from the sources referenced.