Tracking as a Service

This repository contains a set of Tracking models that can be used as a service.

Instructions

1. Launch the server with an interactive job for 4 hours

srun --job-name=TritonTest -C "gpu&hbm80g" -N 1 -G 1 -c 10 -n 1 -t 4:00:00 -A m3443 \
  -q interactive /bin/bash -c "./scripts/start-tritonserver.sh -o triton_ready.txt"

Or, launch the server with a regular batch job for 48 hours

sbatch -G 1 request_servers.sh

2. Run the client:

podman-hpc run -it --rm --ipc=host --net=host --ulimit memlock=-1 --ulimit stack=67108864 \
  -v ${PWD}:/workspace/ -w /workspace \
  -v /global/cfs/cdirs/m3443/data/for_alina:/global/cfs/cdirs/m3443/data/for_alina \
  docker.io/docexoty/tritonserver:latest python models/MetricLearning/2/client.py -i /global/cfs/cdirs/m3443/data/for_alina/all_input_node_features.pt

Models

Supported models are saved in the model_repos directory.

ExaTrkX Models

Python backend

Docker container

podman-hpc build --format docker -f Dockerfile -t docexoty/tritonserver

Install packages.

poetry env use /global/common/software/nersc/pe/conda-envs/24.1.0/python-3.11/nersc-python/bin/python

poetry install --no-root