Train a machine translation model to convert text from one language to another using the encoder-decoder architecture.
Run experiments to reproduce or extend the 2017 Google research paper on neural machine translation architectures.
Build a text summarization model that reads long documents and outputs shorter summaries.
Experiment with a conversational AI model that generates responses to input messages.
| google/seq2seq | facebookresearch/mmf | lucidrains/dalle-pytorch | |
|---|---|---|---|
| Stars | 5,629 | 5,629 | 5,629 |
| Language | Python | Python | Python |
| Setup difficulty | hard | moderate | hard |
| Complexity | 4/5 | 4/5 | 5/5 |
| Audience | researcher | researcher | researcher |
Figures from each repo's GitHub metadata at analysis time.
Academic research codebase, not production-ready, external documentation required and GPU strongly recommended.
This repository contains a general-purpose encoder-decoder framework built on TensorFlow, an AI development library. Encoder-decoder models are a class of AI architecture where one part of the model reads an input sequence and compresses it into a representation, and another part generates an output sequence from that representation. This approach is used for tasks where you need to transform one sequence of tokens into another. The framework was built to support multiple applications: translating text from one language to another, summarizing longer documents into shorter ones, building conversational systems that generate responses to input messages, and generating text descriptions of images. The README is brief and points to an external documentation site for setup and usage details. The code was originally published as the official implementation for a 2017 research paper called Massive Exploration of Neural Machine Translation Architectures, written by researchers at Google. That paper systematically tested a large number of design choices in translation models to understand which architectural decisions matter most. The repository is shared for academic reproducibility and general experimentation rather than as a production-ready library. This is not an officially supported Google product.
A general-purpose TensorFlow encoder-decoder framework from Google, originally built for machine translation research, usable for translation, summarization, and conversation.
Mainly Python. The stack also includes Python, TensorFlow.
Apache 2.0, use freely for any purpose including commercial, as long as you keep the copyright and license notice.
Setup difficulty is rated hard, with roughly 1day+ to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Verify against the repo before relying on details.