NVIDIA's In-Game Inferencing (NVIGI) SDK 1.5 introduces a code agent sample that reduces GPU inference costs in games by having an SLM generate executable code instead of making repeated tool-calling inference calls. The post contrasts tool-calling (multiple inference round-trips per action) with code agents (one inference

10m read time From developer.nvidia.com
Post cover image
Table of contents
Why code agents outshine tool-callingCode agent sample dungeonThe threat model of a code agentChoosing a target languageGet started with NVIDIA In-Game Inferencing SDKJoin us at GDC

Sort: