The latest paper signed by DeepSeek founder Liang Wenfeng was made public. For the first time, the technical details of DeepSeek's agent training sandbox platform DSEC were publicly released. The list of authors exceeded 130, and Liang Wenfeng was on the list. According to the paper, the dSeC platform is very large. One production unit consists of about 160 CPU nodes, 30,000 cores, and 250TB of memory to host PB-level images. In terms of capacity, the dSeC platform serves about 3 million sandboxes in a single day, with a peak concurrent of more than 380,000, and a creation speed of more than 5,000 sandboxes per second. A single training task can pull up at most 32,000 sandboxes at once.

The latest paper signed by DeepSeek founder Liang Wenfeng was made public. For the first time, the technical details of DeepSeek's agent training sandbox platform DSEC were publicly released. The list of authors exceeded 130, and Liang Wenfeng was on the list. According to the paper, the dSeC platform is very large. One production unit consists of about 160 CPU nodes, 30,000 cores, and 250TB of memory to host PB-level images. In terms of capacity, the dSeC platform serves about 3 million sandboxes in a single day, with a peak concurrent of more than 380,000, and a creation speed of more than 5,000 sandboxes per second. A single training task can pull up up up to 32,000 sandboxes at once.