WARNING: PyPI Phishing attack ongoing. Please check that you are on the correct domain - PyPI dot org - before continuing  PyPI

NEMO core package. Necessary for all collections

Navigation

Verified details

These details have been verified by PyPI
Maintainers
Avatar for eharper from gravatar.com eharper Avatar for ko3n1g from gravatar.com ko3n1g Avatar for okuchaiev from gravatar.com okuchaiev

Unverified details

These details have not been verified by PyPI
Project links
Meta
  • License: Apache Software License (Apache2)
  • Author: NVIDIA
  • Maintainer: NVIDIA
  • Tags deep learning , machine learning , gpu , NLP , NeMo , nvidia , pytorch , torch , tts , speech , language
Classifiers
Report project as malware

Project description

NeMo (Neural Modules) is a toolkit for creating AI applications built around neural modules, conceptual blocks of neural networks that take typed inputs and produce typed outputs.

NeMo Core provides the fundamental building blocks for all neural models and NeMo's type system.

Core Principles

NEMO is built around these principles:

  • Neural Module is a block that computes a set of outputs from a set of inputs.
    • Think of it as an abstraction between a layer and a neural network.
    • Examples: pre-trained language model, acoustic model, encoder, decoder, etc.
  • NeMo-based application is a DAG of :class:NeuralModule<nemo.core.neural_modules.NeuralModule> objects connected together via typed input and output ports.
    • :class:NmTensor<nemo.core.neural_types.NmTensor> objects flow between modules from port to port.
    • Lazy execution model. First, user defines activation flow graph. Then, calls action (such as training). Actual computation happen only after action is called.
  • The input and output ports of each neural module and :class:NmTensor<nemo.core.neural_types.NmTensor> objects all have Neural Type - :class:NeuralType<nemo.core.neural_types.NeuralType>.
  • :class:NeuralType<nemo.core.neural_types.NeuralType> the semantics, axis order, and dimensions of a tensor and more.
  • Domain or module-specific Callbacks are used for evaluations and inspecting training performance.

Built to run on GPUs

NeMo Core provides:

  • Mixed-precision training using Tensor Cores on NVIDIA's Volta and Turing GPUs
  • Distributed training
  • Distributed evaluation

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page