XVERSE AI
Agent Recognition AI-Drive General Knowledge Foundation for All Scenarios XVERSE-LLM
Data - Information Processing
Better Data, Less Hallucination
Engineering - High-Efficiency GPU Cluster
Faster Inference, Higher Concurrent Access
Infrastructure - Massive System
Richer Ecosystem, Better Knowledge Enhancement
XVERSE-LLM
All In-House R&D with Superior Performance
Complex distributed system design with a peak computing power utilization rate of 58.5%;
FlashAttention2 accelerated computation, 3D parallelism, and virtual pipelining, enhancing inference computation speed;
Optimization of cluster infrastructure, resource allocation, and training framework, with an effective training rate of 98.6%.
XVERSE-65B
XVERSE-Long
XVERSE-MoE
65BModel Size
2.6TDataset Size
40+Language
16KContext Length
Product Detail
