Skip to content

Personal collection of example scripts and configurations for working with NVIDIA Triton Inference server

Notifications You must be signed in to change notification settings

IMG-PRCSNG/triton-playground

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

TRITON Inference server - Example Configuration and scripts

Personal collection of examples and scripts for working with NVIDIA Triton Inference Server

Pre-requisites

Python backend

Python backend in Triton Inference Server allows us to write custom python functions. It is useful in scenarios where exporting model to formats such as torchscript, onnx, savedmodel etc is unavailable / fails / operators are unsupported outside native execution.

TLDR

# Create execution environment
conda env create -f environment.yml
conda activate triton
conda install conda-pack

# Packs the environment into triton.tar.gz, will take a while
conda pack

mv triton.tar.gz models/
# Model
wget "https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_FPN_3x/137849458/model_final_280758.pkl" -O models/test/1/model.pkl

docker-compose up triton

Once the server is up (should take a while to load the model), you can now use the client

pip install tritonclient[http] Pillow numpy
python3 client.py <IMAGE> --model test --url localhost:8000

Longer version

This example shows how to use Detectron2 Faster RCNN model using the python backend of Triton.

Detectron2 is a Pytorch based DL framework for training Object detection and segmentation models with an easy-to-use API. (Tutorial)

To use models trained with Detectron2 in the Python backend of Triton Server, we have to bundle these dependencies, according to these instructions

TODO - Explainers for

  1. conda pack
  2. Config.pbtxt
  3. model.py
  4. client.py

About

Personal collection of example scripts and configurations for working with NVIDIA Triton Inference server

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages