Wire flash
DeepSeek founder Liang Wenfeng co-authors paper detailing Agent training sandbox platform DSec
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
DeepSeek founder Liang Wenfeng is among over 130 authors of a newly published paper that details the technical specifications of the company's Agent training sandbox platform, DSec (DeepSeek Elastic Compute). According to the paper, a single production unit of the DSec platform comprises approximately 160 CPU nodes, 30,000 cores, and 250TB of memory, hosting petabyte-scale images. The platform demonstrates significant scale, serving about 3 million sandboxes daily with a peak concurrency of over 380,000. It can create more than 5,000 sandboxes per second, and a single training task can launch up to 32,000 sandboxes at once. This marks the first systematic public release of the platform's technical details.
Source report
DeepSeek founder Liang Wenfeng has co-authored a newly published research paper that details the technical architecture of DSec (DeepSeek Elastic Compute), the company's agent training sandbox platform. This marks the first systematic release of the platform's technical specifications.
The paper lists over 130 authors, including Liang Wenfeng.
Key Platform Specifications
- Production unit scale: Approximately 160 CPU nodes, 30,000 cores, and 250 TB of memory
- Storage capacity: Hosts petabyte-level images
- Daily throughput: Serves approximately 3 million sandboxes per day
- Peak concurrency: Exceeds 380,000 concurrent sandboxes
- Creation speed: Over 5,000 sandboxes per second
- Single training task capacity: Can launch up to 32,000 sandboxes simultaneously
Source
domesticEastern