Methods, devices, and storage media for handling computational cost in large model training
By preprocessing and analyzing performance credentials of large model training datasets, reliable performance credentials are generated and computing power costs are processed. This solves the problems of disconnect between billing and value and lack of green guidance in existing technologies, and realizes cost optimization for large model training and reliable billing in the market ecosystem.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SI-TECH INFORMATION TECH CO LTD
- Filing Date
- 2025-12-31
- Publication Date
- 2026-05-26
AI Technical Summary
The existing computing power billing model cannot adapt to the complexity and optimization needs of large model tasks, resulting in a disconnect between billing and value, a lack of green and efficiency guidance, and a lack of credible evidence chains to prove fee reductions, which can easily lead to disputes.
By preprocessing the original large model training dataset, performing performance certificate analysis, generating performance certificates, and processing computing power costs based on the certificates, the system includes an import module, a preprocessing module, a performance certificate analysis module, and a computing power cost processing result module. Blockchain technology is used for evidence storage to ensure the fairness and transparency of the incentive mechanism.
It reduced computing power expenditure, improved the overall utilization quality of large-scale computing power clusters, significantly reduced the total cost of training large models, increased throughput, and built a trustworthy computing power market ecosystem through green computing power incentives and a trustworthy billing mechanism.
Smart Images

Figure CN122089415A_ABST