tensorflow/lingvo

Lingvo

What it solves

Lingvo provides a modular and scalable framework for building neural networks in TensorFlow, with a specific focus on sequence models (sequence-to-sequence modeling).

How it works

It allows developers to build and train models across various modalities. The framework supports high-scale training, including model parallelism for giant language models (e.g., GShard) and integration with CloudTPUs and GPUs. It provides a structured approach to defining models through parameters and tasks, supporting both pip-based installation for standard use and Bazel-based builds for framework development.

Who it’s for

Researchers and engineers building complex neural networks, particularly those working on large-scale sequence models, machine translation, and automatic speech recognition.

Highlights

  • Diverse Model Support: Includes implementations for machine translation, automatic speech recognition, language modeling, 3D object detection (StarNet, DeepFusion), and image recognition (LeNet5).
  • Scalability: Supports giant language models with up to one trillion parameters using GShard and 512-way model parallelism.
  • Hardware Acceleration: Optimized for use with GPUs and CloudTPUs.
  • Flexible Installation: Available via pip for quick start or Bazel for those contributing to the framework core.

Related

  • Project
    tensorforce/tensorforce

    A modular TensorFlow-based deep reinforcement learning framework for research and practice, providing flexible components to build and deploy RL agents.

  • Project
    tensorflow/tensorflow

    An end-to-end open source platform for machine learning that provides tools and libraries for building and deploying ML-powered applications.

  • Project
    tensorflow/gnn

    A TensorFlow library for building and scaling Graph Neural Networks, featuring tools for heterogeneous graph representation and distributed graph sampling.

  • Project
    google/seqio

    A library for processing sequential data that provides a unified framework for task-based datasets, preprocessing, and evaluation for sequence models.

  • Project
    AIStream-Peelout/flow-forecast

    An open-source deep learning framework for time series forecasting, classification, and anomaly detection featuring a wide array of state-of-the-art models.