Info This post is auto-generated from RSS feed Hacker News. Source: DeepSeek open-sources inference optimizations with 60–85% faster generation [pdf]