|
--- |
|
license: mit |
|
datasets: |
|
- opus100 |
|
- un_multi |
|
language: |
|
- en |
|
- ar |
|
--- |
|
M2M100 418M |
|
|
|
M2M100 is a multilingual encoder-decoder transformer model trained for Many-to-Many multilingual translation. The model, originally introduced by researchers at Facebook, demonstrates impressive performance in cross-lingual translation tasks. |
|
For a better understanding of M2M100 you can look into the [paper](https://arxiv.org/abs/2010.11125) and the associated [repository](https://github.com/facebookresearch/fairseq/tree/main/examples/m2m_100). |
|
To further enhance the capabilities of M2M100, we conducted finetuning experiments on English-to-Arabic parallel text. The finetuning process involved training the model for 1000K steps using a batch size of 8. |