Glot500
[ACL 2023] Glot500: Scaling Multilingual Corpora and Language Models to 500 Languages
- 109
- Python
- 7 mo ago
[ACL 2023] Glot500: Scaling Multilingual Corpora and Language Models to 500 Languages
[EACL 2021] Self-training Pretrained LMs for Zero- and Few-shot Arabic Sequence Labeling
An attempt to make Back-Translation differentiable, using probability weighted embeddings for predicted translations in the nucleus of the predicted distribution over target language sentences.
XLM implementation with utilities to process and train on large multi-lingual datasets, with not enough RAM.