Understand this story
Story summary
The English summary is being updated. See the reporting timeline below.
Latest development · Report
Exa Boosts Embedding Throughput by 50% with AWS AI Chips, Indexing Billions of DocumentsDevelopments
1 confirmed developments
- Source:AWS Machine Learning
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 19
AWS Machine LearningFirst-party
Exa Boosts Embedding Throughput by 50% with AWS AI Chips, Indexing Billions of Documents
Exa optimized its embedding infrastructure on AWS Inferentia2 and AWS Trainium2, increasing throughput by approximately 1.5× for large-scale document indexing. It regenerated embeddings for 30 billion high-priority documents and reduced compute and memory overhead through weight layout optimization, autocast, NKI kernels, and a custom SwiGLU MLP kernel, achieving a cosine similarity of 0.999 between its output and the reference results.
AI Trends