Trick to get the 2 DGX Spark TRT LLM setup running (fixes stuck on “preparing”)

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
1

Before deploying the Docker stack, run the following on each Spark:

docker pull nvcr.io/nvidia/tensorrt-llm/release:1.0.0rc3

Then deploy the stack as in the instructions: Try NVIDIA NIM APIs

This fixes the issue of the nodes being stuck on “preparing”.

2

Which issue are you referring to?

The TensorRT LLM playbook does use tensorrt-llm/release:1.0.0rc3