Multi-Model Inference with Triton Inference Server

January 10, 2024

In the modern deep learning era, post-Transformer, there has been exponential growth in AI startups. These startups leverage cutting-edge research to build innovative products for both users and businesses. However, creating and deploying a scalable machine-learning-embedded product that meets end-to-end client requirements is challenging. Challenges include handling streaming data volumes, integrating new models, and achieving extremely low latency. Enter Triton Inference Server, an open-source software enabling the deployment of deep learning models for inference in production environments.

In this blog post, we will explore how Triton is used to deploy a multi-model deep learning architecture. We'll also take a deep dive into understanding the superb features of the Triton Inference Server.

Challenges with Multi-Model Real-Time Inference

Real-time inference of deployed deep learning models is a pretty complex task. Here, we will discuss a few challenges that need to be addressed for successful deployment in various applications.

Low Latency: The performance of the deployed model is measured upon really strict low latency constraints (practically about 200ms) and optimized over high throughput.

Scalability: When we scale our infrastructure to handle a large number of simultaneous requests, it requires load balancing and distributed computing solutions to ensure that the infrastructure can handle varying loads in a cost-effective manner.

Model Complexity: Multiple models can be highly complex, as they need to understand and integrate information from different sources. Training and deploying these models for real-time inference can be resource-intensive.

Resource Management: Managing resources, such as CPU, GPU, memory, and storage, is crucial for real-time inference. Ensuring that resources are allocated efficiently to handle multi-model data can be complex.

Deployment and Maintenance: Deploying and maintaining multi-model real-time systems in the field can be complex, as they often require ongoing updates, maintenance, and monitoring.

In the upcoming section, we will understand the key functionalities and the best features of the Triton Inference Server.

Understanding Triton Inference Server

Figure: Triton Server Architecture with highlighted features (source: Triton Architecture)

Let's have a look at the 7 most powerful features of the Triton Inference Server:

1. Computational Devices Plugin

In a deployment environment, there can be multiple associated models to fulfill a single task or an ensemble of tasks. These models might need different types of compute backends (GPUs/ CPUs). Triton provides support for hosting multiple models on different computational devices such as GPUs and CPUs.

2. Monitoring Feature

When we host a server in production, it is really important that we plug-in monitoring pipelines. It can be model monitoring or infra monitoring. The Triton server has a built-in metric monitoring system, which exports its metric reports using http; hence it can be seamlessly integrated with any other monitoring system like the Prometheus Grafana dashboard which collects these metrics.

3. Multiple Framework Support

The Triton Inference Server provides flexibility to host models built upon different deep learning frameworks such as PyTorch, TensorFlow, ONNX, etc. This feature enables us to seamlessly develop task-specific models and deploy them to serve a client's request.

4. Scheduler

When we use multiple models, each with multiple versions, and each receives a high volume of data input requests, we need an effective orchestrator and scheduler. This functionality is in-built in the Triton Inference Server.

5. HTTP and gRPC

Triton also supports HTTP and gRPC, allowing integration through both HTTP and gRPC requests using different ports.

6. Client Application

Triton has numerous options for SDKs in different languages; Python, C++, Java, which can be used to create client applications in order to interact with the inference server.

7. Model Analyzer

This tool, provided by Triton, offers a range of optimization features through the Triton Model Analyzer. We can think of it as a grid search tool that explores various optimization options and provides us with the most optimized combination of model configurations.

Getting Started

Docker and Triton Installations

!apt-get -qq install
docker pull

You can check the installation of the server as:

Setting Up the Project Repository

Triton expects the models, configurations, version files, etc. in a specific file structure format. These can be stored in a local file system or cloud object store like E2E Solutions. Check out the object storage options on their website:

Triton for Gen AI Inference

In this section, we will deploy and inference a Generative AI model, namely, DCGAN on FASHIONGEN dataset from PyTorch Hub and understand how to use Triton for single model inference. The official Triton Inference Server GitHub repository contains Quick_Deploy examples for different frameworks such as ONNX, TensorFlow, vLLM, and so on.

Figure: Inference pipeline for deployment and generation with DCGAN.

In this section, we will use the single pipeline approach for client-model interaction. Here, we will deploy the pipeline without explicitly breaking apart the model from the pipeline. 

Setting Up the Project Repository

As discussed earlier, we will have to set up our model repository in a Triton's readable format, as provided below:

mkdir -p model_repository/DCGAN_FASHIONGEN/1

After creating the file structure, save the ‘’ file in folder 1, and ‘config.pbtxt’ file in the DCGAN_FASHIONGEN folder, indicating the version of the model to Triton.

   +-- config.pbtxt 
   +-- 1

Now, as we have our file structure ready, we will move towards the PyTorch hub to save our generative ai model ‘DCGAN’.

Preparing the Torchscript Model

We will prepare a file to export the model and save a trace of the 64x64 size image.

!pip3 install torch torchvision torchaudio --index-url
!pip3 install torchvision

import torch
import torchvision.models as models
use_gpu = True if torch.cuda.is_available() else False
model = torch.hub.load('facebookresearch/pytorch_GAN_zoo:hub', 'DCGAN', pretrained=True, useGPU=use_gpu), "")
print("TorchScript DCGAN model saved.")

Now, as we have the model and repository structure ready, let's look at the config.pbtxt model configuration file. The minimum requirements for the configuration file are that you must satisfy the platform (or backend properties), the max_batch_size property, and the input and output tensors of the model.

Setup Model Configuration

platform: "pytorch_libtorch"
max_batch_size : 0
input [
   name: "input__0"
   data_type: TYPE_INT32
input [
   name: "input__1"
   data_type: TYPE_FP32
   dims: [ 32, 120 ]
output [
   name: "output__0"
   data_type: TYPE_FP32
   dims: [ 32, 120]
output [
   name: "output__1"
   data_type: TYPE_FP32
   dims: [ 32, 3 ,64, 64]
   reshape { shape: [ 32, 64, 64, 3 ] }
  • Name: ‘name’ is an optional field, the value of which should match the name of the directory of the model.
  • Backend: This field indicates which backend is being used to run the model. Triton supports a wide variety of backends like TensorFlow, PyTorch, Python, ONNX and more.
  • max_batch_size: As the name implies, this field defines the maximum batch size that the model can support.
  • Input and output: The input and output sections specify the name, shape, datatype, and more, while providing operations like reshaping.

Now that we are ready to launch the server, we will use the freely available pre-built Docker containers from NGC.

Launch the Triton Server

docker run --gpus all --rm -p 8000:8000 -p 8001:8001 -p 8002:8002 -v ${PWD}/model_repository:/models tritonserver --model-repository=/models

Now we have to build a Client, which requires three basic points:

  • It should set up a connection with the Triton Inference Server.
  • It should specify the names of the input and output layers of our model.
  • It should send an inference request to the Triton Inference Server.

First download the dependencies (torchvision) inside our workspace:

docker run -it --net=host -v ${PWD}:/workspace/ bash
pip install torchvision

Set up connection with the client.

client = httpclient.InferenceServerClient(url="localhost:8000")

Specify the names of the input and output layers of the model.

num_images = 32
input_0 = httpclient.InferInput("input__0", dtype="INT32")
input_0.set_data_from_numpy(num_images, binary_data=True)
output_0, _ = httpclient.InferRequestedOutput("output__0", binary_data=True, class_count=48)
input_1 = httpclient.InferInput("input__1", output_0.shape, datatype="FP32")
input_1.set_data_from_numpy(transformed_img, binary_data=True)
outputs = httpclient.InferRequestedOutput("output__1", binary_data=True, class_count=48)

Send an inference request to the server.

results = client.infer(model_name="DCGAN_FASHIONGEN", inputs=[input_0, input_1], outputs=[output_0, output_1])
generated_image = results.as_numpy('output__1')

The model output should look like:

Multi-Model Inference with NVIDIA Triton

In this section, we will understand the workflow for the deployment of multiple models. For this we will take a deeper look at the official docs provided by triton-inference-server GitHub repository.

The agenda for this part is to deploy a pipeline for transcribing a text-from-images model. However, in this section, we will use a break apart pipeline, and will leverage different backends for multiple models’ input-output processing while deploying core models on different framework backends.

Ensemble pipeline: We will divide the problem into two major components: the text detection pipeline and the text recognition pipeline. The Triton Inference Server allows us to deconstruct the model deployment pipeline and build an ensemble model, applying pre- and post-processing steps along with the exported models.

So let's get started with the multi-model deployment – and we will learn some cool features of Triton.

Downloading the Model

  • Text Detection

tar -xvf frozen_east_text_detection.tar.gz

  • Text Recognition


Export to ONNX

  • NGC TensorFlow container environment: docker run -it --gpus all -v ${PWD}:/workspace<>-tf2-py3
  • install tf2onnx: pip install -U tf2onnx
  • Converting OpenCV's EAST model to ONNX:

python -m tf2onnx.convert --input frozen_east_text_detection.pb --inputs "input_images:0" --outputs "feature_fusion/Conv_7/Sigmoid:0","feature_fusion/concat_3:0" --output detection.onnx

Setting Up the Model's Repository

As we have already looked at Triton's way of reading our model, let's now focus on setting up a model repository for these multiple models.

This file structure can be set up in the following manner:

mkdir -p model_repository/text_detection/1
mv detection.onnx model_repository/text_detection/1/model.onnx
mkdir -p model_repository/text_recognition/1
mv str.onnx model_repository/text_recognition/1/model.onnx

Setting Up the Model Configurations

The Triton Inference Server GitHub repository provides the configuration files of text_detection and text_recognition.

Installing and Importing Dependencies

We will need the following dependencies for image processing and HTTP client.

!pip install cv2
!pip install tritonclient
import math
import numpy as np
import cv2
import tritonclient.http as httpclient

Launching the Triton Server

docker run --gpus=all -it --shm-size=256m --rm -p8000:8000 -p8001:8001 -p8002:8002 -v $(pwd)/model_repository:/models

Building a Client Application

As we discussed three key points to build a client, the conceptual guide (of triton repository) has defined a few helper functions that take care of pre- and post-processing steps in our pipeline. You can check out the file. 

Let's recall three golden points for the client application:

  • It should set up a connection with the Triton Inference Server

client = httpclient.InferenceServerClient(url="localhost:8000")

It should specify the names of the input and output layers of our model.

raw_image = cv2.imread("./img2.jpg")
preprocessed_image = detection_preprocessing(raw_image)
detection_input = httpclient.InferInput("input_images:0", preprocessed_image.shape, datatype="FP32")
detection_input.set_data_from_numpy(preprocessed_image, binary_data=True)

It should send an inference request to the Triton Inference Server.

detection_response = client.infer(model_name="text_detection", inputs=[detection_input])

We will repeat the process for text recognition model to finally perform the post-process and print the results:

  • Process responses from the detection model.

scores = detection_response.as_numpy('feature_fusion/Conv_7/Sigmoid:0')
geometry = detection_response.as_numpy('feature_fusion/concat_3:0')
cropped_images = detection_postprocessing(scores, geometry, preprocessed_image)

Create an input object for the recognition model.

recognition_input = httpclient.InferInput("input.1", cropped_images.shape, datatype="FP32")
recognition_input.set_data_from_numpy(cropped_images, binary_data=True)

Query the server.

recognition_response = client.infer(model_name="text_recognition", inputs=[recognition_input])

Process the response from the recognition model.

text = recognition_postprocessing(recognition_response.as_numpy('308'))

Voilà! Now we understand how seamless multi-model deployment using the Triton Inference Server is. Do check out the Triton Inference Server GitHub repository to try out the text detection and recognition deployment yourself.

Latest Blogs
This is a decorative image for: A Complete Guide To Customer Acquisition For Startups
October 18, 2022

A Complete Guide To Customer Acquisition For Startups

Any business is enlivened by its customers. Therefore, a strategy to constantly bring in new clients is an ongoing requirement. In this regard, having a proper customer acquisition strategy can be of great importance.

So, if you are just starting your business, or planning to expand it, read on to learn more about this concept.

The problem with customer acquisition

As an organization, when working in a diverse and competitive market like India, you need to have a well-defined customer acquisition strategy to attain success. However, this is where most startups struggle. Now, you may have a great product or service, but if you are not in the right place targeting the right demographic, you are not likely to get the results you want.

To resolve this, typically, companies invest, but if that is not channelized properly, it will be futile.

So, the best way out of this dilemma is to have a clear customer acquisition strategy in place.

How can you create the ideal customer acquisition strategy for your business?

  • Define what your goals are

You need to define your goals so that you can meet the revenue expectations you have for the current fiscal year. You need to find a value for the metrics –

  • MRR – Monthly recurring revenue, which tells you all the income that can be generated from all your income channels.
  • CLV – Customer lifetime value tells you how much a customer is willing to spend on your business during your mutual relationship duration.  
  • CAC – Customer acquisition costs, which tells how much your organization needs to spend to acquire customers constantly.
  • Churn rate – It tells you the rate at which customers stop doing business.

All these metrics tell you how well you will be able to grow your business and revenue.

  • Identify your ideal customers

You need to understand who your current customers are and who your target customers are. Once you are aware of your customer base, you can focus your energies in that direction and get the maximum sale of your products or services. You can also understand what your customers require through various analytics and markers and address them to leverage your products/services towards them.

  • Choose your channels for customer acquisition

How will you acquire customers who will eventually tell at what scale and at what rate you need to expand your business? You could market and sell your products on social media channels like Instagram, Facebook and YouTube, or invest in paid marketing like Google Ads. You need to develop a unique strategy for each of these channels. 

  • Communicate with your customers

If you know exactly what your customers have in mind, then you will be able to develop your customer strategy with a clear perspective in mind. You can do it through surveys or customer opinion forms, email contact forms, blog posts and social media posts. After that, you just need to measure the analytics, clearly understand the insights, and improve your strategy accordingly.

Combining these strategies with your long-term business plan will bring results. However, there will be challenges on the way, where you need to adapt as per the requirements to make the most of it. At the same time, introducing new technologies like AI and ML can also solve such issues easily. To learn more about the use of AI and ML and how they are transforming businesses, keep referring to the blog section of E2E Networks.

Reference Links

This is a decorative image for: Constructing 3D objects through Deep Learning
October 18, 2022

Image-based 3D Object Reconstruction State-of-the-Art and trends in the Deep Learning Era

3D reconstruction is one of the most complex issues of deep learning systems. There have been multiple types of research in this field, and almost everything has been tried on it — computer vision, computer graphics and machine learning, but to no avail. However, that has resulted in CNN or convolutional neural networks foraying into this field, which has yielded some success.

The Main Objective of the 3D Object Reconstruction

Developing this deep learning technology aims to infer the shape of 3D objects from 2D images. So, to conduct the experiment, you need the following:

  • Highly calibrated cameras that take a photograph of the image from various angles.
  • Large training datasets can predict the geometry of the object whose 3D image reconstruction needs to be done. These datasets can be collected from a database of images, or they can be collected and sampled from a video.

By using the apparatus and datasets, you will be able to proceed with the 3D reconstruction from 2D datasets.

State-of-the-art Technology Used by the Datasets for the Reconstruction of 3D Objects

The technology used for this purpose needs to stick to the following parameters:

  • Input

Training with the help of one or multiple RGB images, where the segmentation of the 3D ground truth needs to be done. It could be one image, multiple images or even a video stream.

The testing will also be done on the same parameters, which will also help to create a uniform, cluttered background, or both.

  • Output

The volumetric output will be done in both high and low resolution, and the surface output will be generated through parameterisation, template deformation and point cloud. Moreover, the direct and intermediate outputs will be calculated this way.

  • Network architecture used

The architecture used in training is 3D-VAE-GAN, which has an encoder and a decoder, with TL-Net and conditional GAN. At the same time, the testing architecture is 3D-VAE, which has an encoder and a decoder.

  • Training used

The degree of supervision used in 2D vs 3D supervision, weak supervision along with loss functions have to be included in this system. The training procedure is adversarial training with joint 2D and 3D embeddings. Also, the network architecture is extremely important for the speed and processing quality of the output images.

  • Practical applications and use cases

Volumetric representations and surface representations can do the reconstruction. Powerful computer systems need to be used for reconstruction.

Given below are some of the places where 3D Object Reconstruction Deep Learning Systems are used:

  • 3D reconstruction technology can be used in the Police Department for drawing the faces of criminals whose images have been procured from a crime site where their faces are not completely revealed.
  • It can be used for re-modelling ruins at ancient architectural sites. The rubble or the debris stubs of structures can be used to recreate the entire building structure and get an idea of how it looked in the past.
  • They can be used in plastic surgery where the organs, face, limbs or any other portion of the body has been damaged and needs to be rebuilt.
  • It can be used in airport security, where concealed shapes can be used for guessing whether a person is armed or is carrying explosives or not.
  • It can also help in completing DNA sequences.

So, if you are planning to implement this technology, then you can rent the required infrastructure from E2E Networks and avoid investing in it. And if you plan to learn more about such topics, then keep a tab on the blog section of the website

Reference Links

This is a decorative image for: Comprehensive Guide to Deep Q-Learning for Data Science Enthusiasts
October 18, 2022

A Comprehensive Guide To Deep Q-Learning For Data Science Enthusiasts

For all data science enthusiasts who would love to dig deep, we have composed a write-up about Q-Learning specifically for you all. Deep Q-Learning and Reinforcement learning (RL) are extremely popular these days. These two data science methodologies use Python libraries like TensorFlow 2 and openAI’s Gym environment.

So, read on to know more.

What is Deep Q-Learning?

Deep Q-Learning utilizes the principles of Q-learning, but instead of using the Q-table, it uses the neural network. The algorithm of deep Q-Learning uses the states as input and the optimal Q-value of every action possible as the output. The agent gathers and stores all the previous experiences in the memory of the trained tuple in the following order:

State> Next state> Action> Reward

The neural network training stability increases using a random batch of previous data by using the experience replay. Experience replay also means the previous experiences stocking, and the target network uses it for training and calculation of the Q-network and the predicted Q-Value. This neural network uses openAI Gym, which is provided by taxi-v3 environments.

Now, any understanding of Deep Q-Learning   is incomplete without talking about Reinforcement Learning.

What is Reinforcement Learning?

Reinforcement is a subsection of ML. This part of ML is related to the action in which an environmental agent participates in a reward-based system and uses Reinforcement Learning to maximize the rewards. Reinforcement Learning is a different technique from unsupervised learning or supervised learning because it does not require a supervised input/output pair. The number of corrections is also less, so it is a highly efficient technique.

Now, the understanding of reinforcement learning is incomplete without knowing about Markov Decision Process (MDP). MDP is involved with each state that has been presented in the results of the environment, derived from the state previously there. The information which composes both states is gathered and transferred to the decision process. The task of the chosen agent is to maximize the awards. The MDP optimizes the actions and helps construct the optimal policy.

For developing the MDP, you need to follow the Q-Learning Algorithm, which is an extremely important part of data science and machine learning.

What is Q-Learning Algorithm?

The process of Q-Learning is important for understanding the data from scratch. It involves defining the parameters, choosing the actions from the current state and also choosing the actions from the previous state and then developing a Q-table for maximizing the results or output rewards.

The 4 steps that are involved in Q-Learning:

  1. Initializing parameters – The RL (reinforcement learning) model learns the set of actions that the agent requires in the state, environment and time.
  2. Identifying current state – The model stores the prior records for optimal action definition for maximizing the results. For acting in the present state, the state needs to be identified and perform an action combination for it.
  3. Choosing the optimal action set and gaining the relevant experience – A Q-table is generated from the data with a set of specific states and actions, and the weight of this data is calculated for updating the Q-Table to the following step.
  4. Updating Q-table rewards and next state determination – After the relevant experience is gained and agents start getting environmental records. The reward amplitude helps to present the subsequent step.  

In case the Q-table size is huge, then the generation of the model is a time-consuming process. This situation requires Deep Q-learning.

Hopefully, this write-up has provided an outline of Deep Q-Learning and its related concepts. If you wish to learn more about such topics, then keep a tab on the blog section of the E2E Networks website.

Reference Links

This is a decorative image for: GAUDI: A Neural Architect for Immersive 3D Scene Generation
October 13, 2022

GAUDI: A Neural Architect for Immersive 3D Scene Generation

The evolution of artificial intelligence in the past decade has been staggering, and now the focus is shifting towards AI and ML systems to understand and generate 3D spaces. As a result, there has been extensive research on manipulating 3D generative models. In this regard, Apple’s AI and ML scientists have developed GAUDI, a method specifically for this job.

An introduction to GAUDI

The GAUDI 3D immersive technique founders named it after the famous architect Antoni Gaudi. This AI model takes the help of a camera pose decoder, which enables it to guess the possible camera angles of a scene. Hence, the decoder then makes it possible to predict the 3D canvas from almost every angle.

What does GAUDI do?

GAUDI can perform multiple functions –

  • The extensions of these generative models have a tremendous effect on ML and computer vision. Pragmatically, such models are highly useful. They are applied in model-based reinforcement learning and planning world models, SLAM is s, or 3D content creation.
  • Generative modelling for 3D objects has been used for generating scenes using graf, pigan, and gsn, which incorporate a GAN (Generative Adversarial Network). The generator codes radiance fields exclusively. Using the 3D space in the scene along with the camera pose generates the 3D image from that point. This point has a density scalar and RGB value for that specific point in 3D space. This can be done from a 2D camera view. It does this by imposing 3D datasets on those 2D shots. It isolates various objects and scenes and combines them to render a new scene altogether.
  • GAUDI also removes GANs pathologies like mode collapse and improved GAN.
  • GAUDI also uses this to train data on a canonical coordinate system. You can compare it by looking at the trajectory of the scenes.

How is GAUDI applied to the content?

The steps of application for GAUDI have been given below:

  • Each trajectory is created, which consists of a sequence of posed images (These images are from a 3D scene) encoded into a latent representation. This representation which has a radiance field or what we refer to as the 3D scene and the camera path is created in a disentangled way. The results are interpreted as free parameters. The problem is optimized by and formulation of a reconstruction objective.
  • This simple training process is then scaled to trajectories, thousands of them creating a large number of views. The model samples the radiance fields totally from the previous distribution that the model has learned.
  • The scenes are thus synthesized by interpolation within the hidden space.
  • The scaling of 3D scenes generates many scenes that contain thousands of images. During training, there is no issue related to canonical orientation or mode collapse.
  • A novel de-noising optimization technique is used to find hidden representations that collaborate in modelling the camera poses and the radiance field to create multiple datasets with state-of-the-art performance in generating 3D scenes by building a setup that uses images and text.

To conclude, GAUDI has more capabilities and can also be used for sampling various images and video datasets. Furthermore, this will make a foray into AR (augmented reality) and VR (virtual reality). With GAUDI in hand, the sky is only the limit in the field of media creation. So, if you enjoy reading about the latest development in the field of AI and ML, then keep a tab on the blog section of the E2E Networks website.

Reference Links

Build on the most powerful infrastructure cloud

A vector illustration of a tech city using latest cloud technologies & infrastructure