[Authored by Liang Wenfeng, DeepSeek Releases New Paper Targeting Large-Scale Agent Training] According to a report by Jinse Finance, the latest paper authored by DeepSeek founder Liang Wenfeng has been made public, systematically unveiling for the first time the technical details of DeepSeek's agent training sandbox platform, DSec (DeepSeek Elastic Compute). The author list includes over 130 contributors, with Liang Wenfeng among them. The paper reveals that the DSec platform is of an enormous scale, with a single production unit comprising approximately 160 CPU nodes, 30,000 cores, 250TB of memory, and hosting petabyte-level images. In terms of capabilities, the DSec platform serves around 3 million sandboxes daily, with peak concurrency exceeding 380,000 and a creation speed of over 5,000 sandboxes per second. A single training task can initiate up to 32,000 sandboxes simultaneously.