Case Study Sharing

Case 1: AI Infrastructure Service Capability

      A certain province’s IDC center is constructing a high-end computing power cluster with 128 8-card H100 computing power servers,

Collaborate with industry, academia, and research institutions to jointly evaluate and issue authoritative reports, creating a benchmark for jointly certified new computing power centers. The main job responsibilities are as follows:

1. Complete full-scale testing of the second phase of AI infrastructure construction
2. Complete deployment and optimization testing of large-scale models in the second phase of AI infrastructure construction
3. Support for cluster computing power operation and maintenance system software in the third phase of AI infrastructure construction

Key technical features:

1. Complete integrated basic computing power testing, cluster platform capability testing, and large-scale model training+inference, training performance, and stability testing
2. Using fully automated testing tools, complete testing can be conducted within one week to verify cluster communication efficiency, MFU, token/s efficiency, etc
3. Tested with authoritative certification from the Institute of Information and Communications Technology and provided with authoritative reports based on test results.

Key performance evaluation data:

1. LLM Training Efficiency Test for H100 GPU Cluster
2. Large scale AI training cluster NCCL testing
3. GPFS cluster storage performance testing

The corresponding test content is shown in the following figure:

The capabilities of the cluster system operation and maintenance management software system are shown in the following figure:

Case 2: AI Industry Application Platform Services

Chinese Dental Association Dental Bot Project

• • Building a Chinese Dental Association and constructing a medical and Q&A intelligent system
  • Build a medical and Q&A large-scale model system
  • Build 1 L1 level local vector knowledge base
  • Build an overall technical framework for intelligent agents
  • Integrate the large model and fine tune it

Implementation technology:

• Implement intelligent agent technology framework
• Construction of Local Vector Knowledge Base in Medical System
• Integrate mainstream industry models
• Implementing LoRA+Prompt Tuning Fine tuning for Large Models
• Building an intelligent agent for the dental association’s medical and Q&A system

The overall technical framework is shown in the following figure:

At the same time, the prototype of AGI development tool box has been completed（ https://octops.ai ）Support SaaS and private deployment

Mechanism Model Service Project of a Research Institution

• Realize the transformation of technology research and development from the traditional model of “experience guided experiments” to a new model of “theoretical prediction and experimental verification”, and construct three AI4S models
  
  • Constructing a molecular chain structure surface performance correlation analysis model
  • Constructing a molecular chain structure mechanical property analysis correlation analysis model
  • Constructing a correlation analysis model between polymerization reaction conditions, polymer yield, and polymer properties

Implementation technology:

• AI4s Basic AI Model
  Implementation of Deep Learning Algorithm+Time Series
• Algorithm+AutoML Algorithm
• Realize surface performance analysis of thin film corona discharge, mechanical performance analysis of thin film, high-throughput data analysis of polypropylene, polymerization conditions and yield, and polymer performance analysis