The Fire-Flyer File System (3FS) is a high-performance distributed file system tailored for AI training and inference workloads. It utilizes modern SSDs and RDMA networks to offer a shared storage layer that simplifies the development of distributed applications. Key features include strong consistency, stateless metadata services, efficient data preparation, random access to training samples, high-throughput checkpointing, and cost-effective KVCache for inference. Performance tests show impressive read throughput and sorting efficiency, making 3FS an optimal choice for large-scale AI tasks.

3m read timeFrom github.com
Post cover image
Table of contents
DocumentationPerformanceCheck out source codeInstall dependenciesBuild 3FSRun a test clusterReport Issues

Sort: