Skip to content

Latest commit

 

History

History
44 lines (30 loc) · 1.05 KB

index.md

File metadata and controls

44 lines (30 loc) · 1.05 KB

THUMT Documentation

THUMT is an open-source toolkit for neural machine translation developed by the Tsinghua Natural Language Processing Group. This page describes the document of THUMT-PyTorch.

Contents

Basics

Prerequisites

  • CUDA 10.0
  • PyTorch
  • TensorFlow-2.0 (CPU version)

Installation

pip install --upgrade pip
pip install thumt

Features

  • Multi-GPU training & decoding
  • Multi-worker distributed training
  • Mixed precision training & decoding
  • Model ensemble & averaging
  • Gradient aggregation
  • TensorBoard for visualization

Walkthrough

We provide a step-by-step walkthrough with a running example: WMT 2018 Chinese-English news translation shared task.

Benchmarks

We provide benchmarks on several datasets. See here.