aws/amazon-sagemaker-examples

Features

• Machine Learning Training - Provides reference implementations for building and optimizing predictive models using custom containers and built-in algorithms.
• SageMaker Example Libraries - Provides a comprehensive collection of Jupyter notebooks for implementing machine learning workflows using Amazon SageMaker.
• Deep Learning Frameworks - Implements model training and hosting using pre-built containers for popular deep learning libraries.
• Large Language Model Fine-Tuning - Demonstrates adapting pre-trained large language models to custom datasets using parameter-efficient fine-tuning.
• Hyperparameter Tuning - Provides workflows for iterative hyperparameter tuning to optimize prediction performance.
• Training and Evaluation Pipelines - Implements reproducible pipelines that chain data preparation, model training, and evaluation stages.
• Hyperparameter Optimization - Implements automated methods for searching and selecting the best configuration parameters for models.
• Model Serving Endpoints - Implements the hosting of trained models behind persistent REST APIs for real-time, low-latency predictions.
• ML Data Preparation - Implements cleaning, labeling, and feature store creation to prepare high-quality data for model training.
• Data Engineering Templates - Provides guides for cleaning, labeling, and transforming large datasets and managing feature stores.
• MLOps Pipeline Automation - Orchestrates the end-to-end ML lifecycle from data processing to model monitoring and lineage tracking.
• Model Endpoint Deployment - Hosts trained models as persistent REST endpoints for real-time requests or via large-scale batch transform jobs.
• ML Model Hosting - Demonstrates practical patterns for hosting real-time REST endpoints and running large-scale batch transform jobs.
• Model Serving Endpoints - Hosts trained models behind persistent endpoints to return low-latency predictions via REST API.
• Automated Model Training - Provides examples of using AutoML to automatically select algorithms and tune hyperparameters.
• Model Lineage Trackers - Provides implementation patterns for tracking the provenance of predictions by recording relationships between datasets and training jobs.
• Spark Integrations - Integrates Apache Spark pipeline stages with machine learning workflows to synchronize data engineering and training.
• Experiment Tracking - Organizes training jobs, models, and endpoints into projects to track relationships and compare results.
• Dataset Batch Inference - Processes an entire dataset through a model in a single job to generate large-scale predictions.
• Feature Stores - Demonstrates the storage and retrieval of curated features for reuse across multiple machine learning models.
• Inference Optimization and Tuning - Improves inference speed and accuracy by combining hardware-specific compilation and hyperparameter tuning.
• Model Evaluation Metrics - Computes accuracy, precision, and recall metrics to assess the performance of trained models.
• Training Algorithms - Implements training using pre-configured, built-in algorithms for common machine learning tasks.
• Fine-tuned Model Deployment - Hosts fine-tuned large language models for real-time inference using specialized containers.
• Containerized Training - Provides templates for packaging custom ML algorithms into Docker containers for managed training.
• Distributed Training Loops - Provides implementations for building and optimizing models using distributed training across multiple devices.
• Annotation Dataset Creation - Provides guides for creating and managing labeling jobs to annotate datasets using human review and active learning.
• Hardware-Aware Compilers - Shows how to optimize deep learning models for specific target hardware to reduce memory and increase speed.
• Model Explainability - Measures feature contributions to model outputs to ensure transparency and identify prediction bias.
• Model Performance Selection - Uses automated machine learning to handle feature selection and model generation based on dataset characteristics.
• Model Training Packaging - Demonstrates how to package and import external training logic and pre-trained models into managed environments.
• Reinforcement Learning - Provides examples for training agents to make sequential decisions through interaction with simulated environments.
• Reinforcement Learning Training - Implements reinforcement learning frameworks to train agents that solve complex problems via iterative feedback.
• Training Dataset Processing - Transforms and cleans large datasets using distributed computing tools to prepare high-quality features for training.
• Training Script Generators - Shows how to execute user-provided training scripts with automatic data loading and checkpointing.
• Data Processing - Executes data preprocessing and feature transformation workloads using standard data science frameworks.
• Distributed Data Processing - Runs distributed preprocessing and feature transformation workloads using containerized tools to prepare large datasets.
• Interactive Notebook Environments - Utilizes pre-configured notebook instances for exploratory data analysis and iterative workflow development.
• Managed Training Environments - Utilizes managed compute clusters with automatic scaling to run machine learning training jobs.
• Compute Orchestration Engines - Demonstrates the allocation and scaling of virtualized infrastructure to run training and processing jobs automatically.
• Model-to-Image Packaging - Demonstrates how to package training algorithms and inference code into portable container images for consistent execution.
• Custom Container Images - Hosts pre-trained models or custom algorithms in production endpoints using custom Docker containers.
• ML Lifecycle Orchestration - Automates the machine learning lifecycle by chaining processing and evaluation steps into reproducible pipelines.
• MLOps Guides - Provides reference workflows for automating the machine learning lifecycle, including monitoring and lineage tracking.
• Model Performance Monitoring - Tracks model drift and analyzes data bias in production to maintain long-term accuracy.
• Deployment and Serving - Notebooks and workflows for deploying models on AWS infrastructure.

Star-Verlauf