#119119th of 242 in ML frameworks and libraries

litert

Google's on-device ML runtime, successor to TensorFlow Lite

Stars
3.6k
License
Apache-2.0
Last commit
Nov 2026
Last release
Oct 2026
Language
C++
Stars, last 29 weeks+1,349
25 Apr 2026: 2,251 stars2 May 2026: 2,293 stars9 May 2026: 2,356 stars16 May 2026: 2,390 stars23 May 2026: 2,427 stars30 May 2026: 2,463 stars6 Jun 2026: 2,497 stars13 Jun 2026: 2,536 stars20 Jun 2026: 2,558 stars27 Jun 2026: 2,580 stars4 Jul 2026: 2,605 stars11 Jul 2026: 2,907 stars18 Jul 2026: 3,123 stars25 Jul 2026: 3,191 stars1 Aug 2026: 3,220 stars8 Aug 2026: 3,258 stars15 Aug 2026: 3,292 stars22 Aug 2026: 3,313 stars29 Aug 2026: 3,334 stars5 Sept 2026: 3,362 stars12 Sept 2026: 3,385 stars19 Sept 2026: 3,411 stars26 Sept 2026: 3,443 stars3 Oct 2026: 3,462 stars10 Oct 2026: 3,478 stars17 Oct 2026: 3,496 stars24 Oct 2026: 3,522 stars31 Oct 2026: 3,556 stars7 Nov 2026: 3,583 stars14 Nov 2026: 3,600 stars

2,251 on 25 Apr 20263,600 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

LiteRT is a C++ runtime that runs .tflite models on Android, iOS, Linux, macOS, Windows, web and IoT targets. Its Compiled Model API selects accelerators automatically across CPU (XNNPACK), GPU (ML Drift) and vendor NPUs. It continues TensorFlow Lite, which is now in maintenance mode, and pairs with separate projects for PyTorch conversion, quantization and LLM inference (LiteRT-LM).

Who it is for: Mobile, embedded and web developers shipping on-device ML models

Strengths

  • Compiled Model API picks accelerators automatically, no manual delegate setup
  • NPU support from several vendors: Qualcomm, MediaTek, Google Tensor, Intel, Broadcom, S.LSI
  • Covers Android, iOS, Linux, macOS, Windows and web (WebGPU/WASM via LiteRT.js)
  • Conversion path from PyTorch, TensorFlow and JAX, plus a separate quantizer

Weaknesses

  • Embedded runtime library, not a server; no HTTP API for serving models
  • LLM inference lives in a separate repo, LiteRT-LM
  • Apple Neural Engine, WebNN and Raspberry Pi NPU support are marked coming soon
  • TensorFlow Lite Interpreter API is deprecated; existing code needs migration to V2

What it needs

  • GPU optional
  • Needs Bazel or CMake (build from source), XNNPACK
  • Models: .tflite, .litertlm, PyTorch (via LiteRT Torch), TensorFlow, JAX

What changed

  1. v2.2.0 ↗ (opens in a new tab)Security fix

    “Added integer overflow checks to convolution, reshape and pad-like CPU kernels,”

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in ML frameworks and libraries

See all 242
Also in ML frameworks and libraries
RankProjectScore
1transformersModel-definition library for pretrained text, vision, audio and multimodal models169.3k starsApache-2.094 out of 100
2ultralyticsPython library and CLI for training and running YOLO vision models63.5k starsAGPL-3.094 out of 100
3pytorch-image-modelsPyTorch library of image models, pretrained weights and training scripts37.3k starsApache-2.094 out of 100
#4pytorchPython tensor library with GPU support and tape-based autograd105.8k starscustom license93 out of 100
#5supervisionPython toolkit for loading, annotating and counting computer vision detections Live demo ↗ (opens in a new tab)51.7k starsMIT92 out of 100