Ant Group Co, a Chinese financial services company backed by billionaire Jack Ma, is using semiconductors made in China for cutting costs on training artificial intelligence (AI) models by 20 per cent.
The results attained were even similar to those from Nvidia Corp. chips like the H800, according to the report.
Ant is using its so-called Mixture of Experts machine learning approach to train the models. Its still using Nvidia for AI development, but is now relying mostly on alternatives including from Advanced Micro Devices Inc. and other Chinese chips for its latest models.
This marks Ant’s entry into the AI race between Chinese and US companies which has accelerated since DeepSeek demonstrated how capable models can be trained for far less than the billions invested by OpenAI and Alphabet Inc.’s Google.
Ant even published a research paper this month, claiming that its models at times outperformed those of Meta in certain benchmarks.
This comes at a time when Mixture of Experts (MoE) models are becoming a popular option since it makes the process cheaper and efficient by dividing tasks into smaller sets of data, like having a team of specialists who each focus on a segment of a job.
Ant has said that it cost about 6.35 million yuan ($880,000) to train 1 trillion tokens, but its optimized approach would cut that down to 5.1 million yuan using lower-specification hardware.
The company plans to use this in the large language models it has developed, Ling-Plus and Ling-Lite, for industrial AI solutions in the health care and finance sector, according to the report.
Source: https://www.hindustantimes.com/
