Here at, https://youtu.be/6mCFzDatGGc?t=1732 this time stamp, the actions functionality was explained. How to get the code base for this or any references?
moving this thread to the Jetson Projects forum
Hi @sanujen.20, thanks for your question! That example was just a manual test in text-generation-webui to see if the model could output the code required to intelligently call the plugins (which it seems to do pretty reliably, and Llava-1.5 VLM added this ability too with constrained JSON output). I’ve not yet integrated invoking the plugins/actions into our Jetson codebase for LLM agents yet, but getting close!
For now, you can probably explore this stuff using other agent frameworks like langchain or similar (which we provide a container for too. The difference with my demos is that they are all geared for interactivity and low-latency for realtime conversations/ect, but if you can relax those requirements for your application there are likely existing solutions out there already that could be run (albeit with additional overhead probably)