https://github.com/mrtj/lstm-translator
Word-level translator using LSTMs
https://github.com/mrtj/lstm-translator
Last synced: about 2 months ago
JSON representation
Word-level translator using LSTMs
- Host: GitHub
- URL: https://github.com/mrtj/lstm-translator
- Owner: mrtj
- Created: 2022-07-01T16:46:15.000Z (about 4 years ago)
- Default Branch: master
- Last Pushed: 2022-07-02T10:38:30.000Z (about 4 years ago)
- Last Synced: 2025-03-06T06:19:26.150Z (over 1 year ago)
- Language: Jupyter Notebook
- Size: 91.8 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Word-level translator using LSTMs
This notebook shows how to train with Amazon SageMaker an NLP model that translates from English to a series of languages. The list of languages is limited by the available training data.
The model uses an LSTM based encoder / decoder architeecture, inspired by https://github.com/hlamba28/Word-Level-Eng-Mar-NMT.
The training time depends on the dataset available. Typically an epoch is trained in 30 minutes for small datasets (for example, Hungarian) and 1.5 - 2 hours for big datasets (for example, Italian). With 15-20 epochs you can already get an acceptable result.
The dataset was created from a curated list of anki flash cards (https://ankiweb.net/shared/decks/). Each translated sentences is annotated with CC license string.