A snippet which demonstrates a potential bug in the triton server
SOLVED: I was using old version of triton server. The issue is resolved in the latest version nvcr.io/nvidia/tritonserver:23.12-py3
docker compose build && docker compose down && docker compose up
Note: docker compose down - resolves the issue with error "ln: failed to create symbolic link '/opt/tritonserver/lib/libnvidia-ml.so.1': File exists" after restarting the server
Setup environment (example using conda)
conda create --name=triton-bug python==3.11
conda activate triton-bug
pip install -r requirements.txt
Open a new terminal and run the client
python run_client.py