Deploy ML Model in Production with FastAPI and Docker

Explore four ml deployment types—batch, streaming, real-time, and edge devices—and compare their advantages, disadvantages, latency implications, and resource trade-offs for production models.

Explore ML infrastructure and integration tools across data versioning, processing, labeling, training, deployment, and monitoring to implement end-to-end MLOps pipelines.

Discover the benefits of deploying machine learning models, including automation, real-time decision making, monitoring, scalability, and improved accuracy through feedback, plus packaging and container deployment for cloud-wide use.

Only 13% of ML models reach production; this lecture covers deployment challenges, from data quality and model quality to security and compliance, and building pipelines with AWS S3.

Orchestrate data ingestion, quality checks, preparation, and model training with online testing; manage a production model registry and serve predictions via an http API on EC2, with S3 and GitHub.

Learn how to use A/B testing to compare model versions in production, allocate traffic between models A and B, monitor performance, and select the superior model based on conversions.

Learn to identify and mitigate bias and security risks in ML models before production. Navigate regulatory compliance such as GDPR and HIPAA while deploying from ideation to production.

Learn aws basics for ml deployment with s3 and ec2. Run dockerized fastapi services on ec2 to download data and models from s3, and upload updates after feedback.

Explore the EC2 dashboard to understand instances, images, volumes, and security groups, and plan production deployment with load balancing and cost optimization for real-time API hosting.

Estimate ml ops production costs with the AWS calculator by configuring EC2 on-demand t3 nano, S3 storage, and data transfer in the North Virginia region.

Launch a Windows EC2 instance, estimate cost, select t2 micro with 30 gb EBS, create a key pair, configure VPC and security group, connect via RDP.

Learn how to create an Ubuntu EC2 instance, set up a key pair and security group for SSH access, connect via public IP, and plan production access controls.

Master AWS EC2 basics with Boto3 and AWS CLI, create key pairs and t2.micro instances, configure security groups, and learn code-driven automation for production with start, stop, and delete operations.

Learn to set up AWS credentials for programmatic access with the AWS CLI and boto3. Configure IAM users, access keys, and secret keys, then access EC2 and S3 from Python.

Create and manage AWS EC2 key pairs with Boto3, generate and save a PEM file, then describe instances and prepare for secure SSH access.

Learn to launch an amazon ec2 instance with boto3, configuring ami id, min/max counts, instance type, key pair, vpc, and storage, while describing and launching instances.

Connects to an EC2 instance and tests connectivity by adjusting inbound rules in the default security group, demonstrating how to open traffic from external networks for ML deployment.

Create a security group with Boto3, describe existing security groups, and attach the new group to an EC2 instance, then configure inbound and outbound rules.

Configure a custom security group's inbound rules with Python and boto3, adding a tcp rule for port 22 and a cidr range to allow ssh access to an EC2 instance.

Attach and detach security groups to an EC2 instance using python and boto3 by describing the instance and modifying its group IDs for production deployments.

Learn how to manage EC2 instances with Boto3 by starting, stopping, and deleting them, and implement a status monitoring loop that waits for instance state changes.

Learn to control S3 with the boto3 sdk in python to automate model uploads and downloads, enabling CI/CD for production deployments and bucket and folder management.

Learn to use boto3 with Python to manage S3 services, list buckets, and create new buckets for automated ml ops workflows in AWS.

Learn to upload files to an S3 bucket using the boto3 sdk, including handling local paths, object naming, and renaming files for production deployment of ml models.

Learn how to list all objects in an s3 bucket with boto3 list_object_v2, access the bucket contents, and print file keys like index.html and some file.html.

Download S3 files to a local system using boto3, create local folders, manage bucket and object names, and prepare for downloading full directories for production ML workflows.

Upload an entire directory to S3 using Python and boto3, walking the folder with os.walk, forming S3 prefix and key, and normalizing paths to forward slashes for uploads.

Discover how to download an entire S3 directory to a local path using boto3 paginator and list objects v2, organizing files for production deployment of ML models and data.

Learn to delete all files in an s3 bucket with boto3, listing objects and deleting them before removing the bucket in a machine learning ops pipeline with fastapi and docker.

Explore sentiment classification using a tiny Bert model to balance accuracy and production resource constraints, and learn to build, tokenize, save, and deploy models with Hugging Face transformers to S3.

Explore sentiment classification on IMDb reviews with a tiny BERT model and transformer architectures, using NSP and MLM pre-training plus fine-tuning for binary positive or negative sentiment.

Load imdb sentiment data with huggingface datasets, convert a pandas dataframe to a dataset, split into train and test sets for model training with transformers in production.

Use a transformers auto tokenizer to produce input IDs, attention masks, and token type IDs for a tiny bert model, with padding and truncation for fixed-length batches.

Implement a model evaluation function for sequence classification using the Hugging Face evaluate library, install it with pip, load the accuracy metric, and compute predictions against references.

Build a transformer-based sentiment classifier by adding a classification head to a tiny four-encoder model, using auto model for sequence classification and training with a trainer.

Save a trained tiny bert sentiment analysis model with trainer.save_model, and prepare to upload it to S3. Use a text classification pipeline to run predictions and view probabilities.

Push your trained model to AWS S3 with boto3, creating or reusing buckets, uploading files or folders, and adding production-ready checks using IAM credentials.

Create and organize an S3 bucket structure for production models by adding an ML models folder and data folders, then upload the model directory to S3 with a script.

Learn to classify disaster tweets with supervised learning, tokenize data with transformers, train and assess models using trainer and training arguments, and deploy via S3 in an end-to-end ML pipeline.

Explore building a multitask disaster tweet classification pipeline using Bert and tiny Bert, leveraging real-time Twitter data, data augmentation, and model distillation, with benchmarking to select the right model.

Tokenize disaster tweet data using a tiny Bert model with auto tokenizer, and prepare input IDs, token type IDs, and attention masks for model training and evaluation.

Construct and evaluate a binary text classifier by adding a classification head to a foundation transformer, apply softmax for cross-entropy training, and use a trainer to evaluate accuracy.

Build an ml ops pipeline, train and save a tiny bert disaster tweet model with transformers and hugging face, achieving 81% accuracy, then load it for inference in production.

Push your trained model to aws s3, organizing it under a ml models folder and using a reusable boto3 script to upload files automatically as part of the ml pipeline.

Explore image classification with a vision transformer, using a Hugging Face train dataset, evaluate with classification reports and confusion matrices, and deploy to AWS S3 serving a REST API.

Explore image classification with human pose data from a hugging face dataset, train on the train split, and deploy multiple models to S3 as part of an end-to-end ml pipeline.

Create label-to-id and id-to-label mappings, extract features and labels from the train split, and convert IDs to human-readable labels for production deployment.

Learn image preprocessing for vision transformers using PyTorch, including automatic resizing, normalization, and tensor conversion with an auto image processor and a ViT pretrained checkpoint.

Learn image preprocessing to resize inputs to 24×24 using the shortest edge with image processor logic. Build a transform pipeline with random resized crop, to tensor, and normalize during training.

Import an auto image classification model, configure the trainer with training arguments, and train a ViT model for five epochs, evaluating and saving the best model.

Evaluate the model on the test dataset, achieving 77% accuracy with 15 classes. Use a classification report and confusion matrix to analyze results and prepare production-ready predictions from logits.

Compute the classification report and confusion matrix to reveal 78% accuracy across 15 classes and misclassifications like calling, texting, and listening to music, then plan uploading the model to S3.

Process real images for pose prediction by loading a fine-tuned vit model, using an image processor and pipeline to classify poses from URLs or images.

Push the trained model to AWS S3 using boto3, reusing existing code, saving to the local path vit human pose classification, and preparing for end-to-end production deployment.

Introduce Streamlit for server-side dashboards, cover basic components, and deploy a Streamlit app while downloading a model from S3 to the Streamlit server and adding a loading spinner.

Explore how to deploy a machine learning model with a Streamlit UI, including building a sentiment-analysis app, downloading models from S3, and testing interactions with text input and prediction buttons.

Learn the basics of Streamlit by coding along, importing libraries, and building a simple app with a title, header, subheader, and images, then run it with streamlit run.

Master streamlit basics by displaying text and headings with st.text and st.text, enabling user input via st.text_input and st.text_area, and rendering markdown with st.markdown.

Demonstrates building interactive Streamlit buttons, triggering events with st.button, and conditionally displaying info, errors, warnings, and toast messages to enhance user feedback.

Upload and display images in a streamlit app from urls or local files with adjustable width, and use checkbox to toggle visibility; explore slider, spinner, radio, and select box.

Explore Streamlit widgets like checkbox, radio button, select box, and multi-select to let users choose a single or multiple models, including NLP, image, and audio models.

Explore Streamlit basics by using a slider to set a threshold for classification, balancing precision and recall, and learn to show a spinner during model download in a Streamlit app.

design and deploy a Streamlit app that downloads a sentiment-analysis model from S3 and classifies user reviews with a predict button, showcasing an end-to-end ML pipeline.

Create a streamlit app that downloads a ml model with a spinner, accepts reviews via a text area, and performs sentiment analysis with a transformers pipeline.

Upload and deploy an ml model streamlit app by creating a GitHub repository, configuring gitignore and requirements.txt, and pushing code to Streamlit Share for hosting on the Streamlit server.

Upload your ML model deployment code to GitHub, connect Streamlit Share, configure secrets and credentials, deploy, test predictions, and share a public app while noting production options like EC2.

Learn to deploy an ml model on AWS EC2 by reconfiguring code from EC2 and S3 sections and setting up security groups. Automate start/stop workflows for scalable APIs.

Learn to automate ML model deployment on AWS EC2 using boto3, including one-time setup and daily operations for creating, starting, stopping, and securing instances with key pairs and security groups.

Automate aws ec2 instance lifecycle to start, stop, and terminate, retrieve instance IDs, update security groups, and fetch the public IP for accessing your Streamlit app in production.

Automate deploying a deep learning model on ec2 with a deep learning ami that includes PyTorch, configure volume, and open ports 2280 and 8501.

Terminate an EC2 instance using a Boto3 script, observe status and changing public IP, and contrast code-based management with a fast console approach for development and production readiness.

Launch a fast, production-ready EC2 server with Ubuntu deep learning PyTorch image, then configure storage, security group, and key pair for quick deployment.

Connect to your deep learning ami instance via browser or ssh, using a pem file and the ubuntu user, to access preinstalled python, conda, and cuda.

Connect VS Code to your EC2 server via remote SSH using your PEM file, edit remote files in an ML ops folder, and monitor resources with htop.

Connect your server to GitHub by generating an SSH key, adding it to the SSH agent and your GitHub account, then clone the deployment repo via SSH from VS Code.

Learn to deploy a Streamlit app on AWS EC2, configure AWS CLI and S3 access, install dependencies from requirements.txt, and run the app. Enable automatic restart on reboot for production.

Create an IAM role with S3 full access and attach it to your EC2 instance to access S3 without credentials, then verify with aws s3 ls.

Learn to assign a permanent elastic IP to an EC2 instance so the IP stays fixed. Attach the EC2 S3 full access IAM role for production.