LiteRT: High-Performance On-Device Machine Learning Framework  |  Google AI Edge  |  Google for Developers
Skip to main content
/
English
Deutsch
Español
Français
Indonesia
Português – Brasil
Русский
中文 – 简体
日本語
한국어
Sign in
Google AI Edge
LiteRT
Android
Desktop
Web
LiteRT-LM
MediaPipe
MediaPipe Solutions
MediaPipe Framework
Model Explorer
AI Edge Portal
Google Tensor SDK
Apps
AI Edge Gallery
AI Edge Eloquent
MediaPipe Demos
API Reference
AI Edge
LiteRT
More
LiteRT-LM
MediaPipe
More
Model Explorer
AI Edge Portal
Google Tensor SDK
Apps
More
API Reference
Home
Overview
Migrating from TensorFlow Lite
LiteRT CLI
Overview
Installation
Common Commands
Troubleshooting & Resources
GenAI Deployments
Overview
Convert PyTorch GenAI models
Run LLMs using LiteRT-LM
Model Conversion & Optimization
Overview
Convert PyTorch models
Overview
Convert PyTorch GenAI models
Convert TensorFlow models
Overview
Use pre-trained models
Convert TensorFlow models
Add Signatures
Model compatibility
Overview
Select operators
Select operators Allowlist
Advanced
Fused operators
Operator versions
RNN models
Optimize models
Overview
Post-training quantization
Post-training dynamic range quantization
Post-training integer quantization
Post-training float16 quantization
Post-training integer quantization with int16 activations
Quantization specification
Inspecting quantization errors
Add model metadata
Overview
Metadata Writer API
Design and build models
Overview
Performance best practices
On-device training
Convert JAX models
Overview
Inference & Hardware Acceleration
Overview
GPU acceleration
NPU acceleration
Overview
Run LLMs using LiteRT-LM
Google Tensor
Intel
Qualcomm
MediaTek
Samsung
Creating a new accelerator
Implementing a Compiler Plugin
Implementing the Dispatch API
Accelerator Test Suite (ATS)
Benchmark & Profiling
Benchmark CompiledModel API
Benchmark Interpreter API
Create custom operators
Custom ops of CompiledModel API
Custom ops of Interpreter API
Run on Android
Overview
Run with CompiledModel API (accelerated on GPU/NPU)
Kotlin API
C++ API
Use prebuilt C++ library
Hardware acceleration
GPU acceleration
NPU acceleration
Google Tensor
MediaTek NPU
Qualcomm NPU
Intel NPU
Run with Interpreter API
Google Play services runtime
Overview
Java API
C API
Hardware acceleration
Acceleration service
GPU with Interpreter API
GPU with C/C++ API
NPU delegates
Overview
Qualcomm NPUs for Mobile AI Development
Development tools
Models with metadata
Overview
Generate model interfaces
Customize data input and output
Run on iOS / macOS
Run with CompiledModel API (accelerated on GPU)
C++ API
Use prebuilt C++ library
Run with Interpreter API
Swift API
Core ML delegate
GPU delegate
Run on Web with LiteRT.js
Run with CompiledModel API (accelerated on GPU)
Overview
JavaScript API
Run on Desktop (Linux, Windows)
Run with CompiledModel API (accelerated on GPU)
C++ API
Use prebuilt C++ library
Python API
Run on Embedded & IoT
Run with CompiledModel API
C++ API
Use prebuilt C++ library
Run with Interpreter API
Overview
Get started
Linux-based devices with Python
Understand the C++ library
Build and convert models
Build from Source
Build Compiled Model API
Build with CMake
Build Interpreter API
Build for Android
Build for iOS
Build for Linux-based IoT
Build with CMake
Overview
Cross compilation for ARM
Build Python Wheel
Reduce binary size
Libraries and tools
Task Library
Overview
ImageClassifier
ObjectDetector
ImageSegmenter
ImageEmbedder
ImageSearcher
NLClassifier
BertNLClassifier
BertQuestionAnswerer
TextEmbedder
TextSearcher
AudioClassifier
Customized API
Model Maker
Overview
Images & video
Image classification
Object detection
Text
Text classification
BERT question & answer
Text search
Audio
Audio classification
Speech recognition
Android
Desktop
Web
MediaPipe Solutions
MediaPipe Framework
AI Edge Gallery
AI Edge Eloquent
MediaPipe Demos
Introducing Google AI Edge Portal: Benchmark Edge AI at scale. Sign-up to request access during private preview.
Home
Products
Google AI Edge
LiteRT
Send feedback
Stay organized with collections
Save and categorize content based on your preferences.
LiteRT is Google's on-device framework for high-performance ML & GenAI deployment on edge platforms.
Efficient conversion, runtime, and optimization for on-device machine learning.
Built on the battle-tested foundation of TensorFlow Lite
LiteRT isn't just new; it's the next generation of the world's most widely deployed machine learning runtime. It powers the apps you use every day, delivering low latency and high privacy on billions of devices.
Trusted by the most critical Google apps
100K+ applications, billions of global users
LiteRT Highlights
Cross Platform Ready
Unleash GenAI
Simplified hardware acceleration
Multi-framework support
Deploy via LiteRT
Streamline your deep learning workflow from training to on-device deployment.
1.Obtain a model
Use .tflite pre-trained models or convert PyTorch, JAX or TensorFlow models to .tflite.
Explore models
Learn about conversion
2.Optimize
Use the LiteRT optimization toolkit to quantize your models post-training.
Explore optimization
3.Run
Deploy your model with LiteRT and pick the optimal accelerator for your app.
View deployment targets
Choose Your Development Path
Use LiteRT to deploy AI anywhere—from high-performance mobile apps to resource-constrained IoT devices.
sync
Existing TFLite User
Transitioning to LiteRT to leverage enhanced performance and unified APIs across platforms (Android, Desktop, Web).
camera
BYOM : Bring your own Models
Have a PyTorch model, looking to implement on-device vision or audio experiences.
auto_awesome
Deploying Generative AI Models
Creating sophisticated on-device chatbots using optimized open-weight GenAI models like Gemma or another open-weight model with LiteRT-LM.
developer_board
[Advanced] Model Expert
Authoring custom models or performing deep hardware-specific CPU/GPU/NPU optimizations for peak performance.
Samples, models, and demo
See LiteRT sample app on GitHub
Complete, end-to-end sample apps.
See sample apps
See genAI models
Pre-trained, out-of-the-box Gen AI models.
Go to HuggingFace
See demos - Google AI Edge Gallery App
A gallery that showcases on-device ML/GenAI use cases using LiteRT.
Open Play Store
Blogs and Announcements
Stay up to date with the latest announcements, technical deep dives, and performance benchmarks from the LiteRT team.
Explore more blogs
LiteRT.js, Google's high performance Web AI Inference
LiteRT.js, Google's high-performance web AI inference library for running models using WebAssembly, WebGPU, and WebNN.
LiteRT unlocks Core Ultra NPU performance for AI PC
Learn how Google and Intel integrate LiteRT with OpenVINO to offload AI inference to Intel NPUs for high-performance AI PCs.
Edge AI from the Trenches: A practical guide to LiteRT
Community insights on untangling LiteRT vs LiteRT-LM and deploying real-world LLMs on edge devices.
Accelerating on-device AI: A look at Arm and Google AI Edge optimization
Arm Scalable Matrix Extension 2 (SME2) and Google AI Edge accelerate on-device generative AI using LiteRT and XNNPACK.
Building real-world on-device AI with LiteRT and NPU
Learn how industry leaders build real-world, high-performance on-device AI applications using LiteRT and NPUs.
Bring state-of-the-art agentic skills to the edge with Gemma 4
Deploy agentic and multi-step planning capabilities entirely on-device with the new Gemma 4 family and LiteRT.
LiteRT: The universal framework for on-device AI
Google's unified on-device ML framework, evolving from TFLite for high-performance deployment.
MediaTek NPU and LiteRT: Powering the next generation of on-device AI
Expanding NPU acceleration support to MediaTek chipsets for high-efficiency AI.
Unlocking Peak Performance on Qualcomm NPU with LiteRT
Unlocking breakthrough performance for generative AI on Qualcomm Neural Processing Units.
LiteRT: Maximum Performance, Simplified
Introducing the CompiledModel API for automated hardware selection and async execution.
On-device GenAI in Chrome, Chromebook Plus, and Pixel Watch with LiteRT-LM
Deploy language models on wearables and browser-based platforms using LiteRT-LM.
Google AI Edge small language models, multimodality, and function calling
Latest insights on RAG, multimodality, and function calling for edge language models
Join the Community
LiteRT GitHub Community
Contribute directly to the project and collaborate with core developers.
Hugging Face Hub
Access optimized open-weight models on the Hugging Face Hub.
lightbulb
LiteRT Collaboration and Feature Intake
Submit feature requests and Collaboration with LiteRT team.
Start Your LiteRT Journey
Ready to take your on-device ML to the next level? Explore the documentation and start building today.
Explore the Docs
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2026-07-17 UTC.
Need to tell us more?
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Missing the information I need","missingTheInformationINeed","thumb-down"],["Too complicated / too many steps","tooComplicatedTooManySteps","thumb-down"],["Out of date","outOfDate","thumb-down"],["Samples / code issue","samplesCodeIssue","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2026-07-17 UTC."],[],[]]
Connect
Blog
Bluesky
Instagram
LinkedIn
X (Twitter)
YouTube
Programs
Google Developer Program
Google Developer Groups
Google Developer Experts
Accelerators
Google Cloud & NVIDIA
Developer consoles
Google API Console
Google Cloud Platform Console
Google Play Console
Firebase Console
Actions on Google Console
Cast SDK Developer Console
Chrome Web Store Dashboard
Google Home Developer Console
Android
Chrome
Firebase
Google Cloud Platform
Google AI
All products
Terms
Privacy
Manage cookies
English
Deutsch
Español
Français
Indonesia
Português – Brasil
Русский
中文 – 简体
日本語
한국어