AI and neural networks
DeepSeek releases the powerful Fire-Flyer open source file system

DeepSeek AI made its parallel file system Fire-Flyer Fire System (3FS) fully open source this week as part of its Open Source Week event. This Chinese company claims that 3FS can achieve an aggregate read throughput of 7.3 TB/s in its data clusters, where it has been using 3FS to organize servers since 2019. 3FS is a Linux-based parallel file system designed for AI-HPC operations, where GPU nodes continuously access multiple storage servers to train large language models (LLMs). The main difference between 3FS and other file systems is the prioritization of random read rates and ignoring read caching. When training models […]
The article DeepSeek Releases Powerful Open Source Fire-Flyer File System was first published at ITZine.ru.
Subscribe
Login
0 Comments
Oldest