Based on in-house AI chips and computing platforms, the company successfully completed the entire training process of a hundred-billion parameter financial LLM in 48 hours, setting a new industry record.

This achievement is attributed to the high-performance tensor computing capabilities of in-house chips, optimized distributed training framework, and intelligent task scheduling system. Compared to traditional GPU solutions, training time was reduced by 60% and computing costs by 40%.

The trained financial LLM performs excellently in scenarios such as risk assessment, investment analysis, and intelligent customer service, and has been piloted in multiple financial institutions with strong client endorsement.