Hi, I’m fine-tuning the google/mt5-base model for many2many translation. The issue is that I do not see any language tag available in the model config (facebook/mbart-large-50 has these included). Normally when generate I would use the forced_bos_token_id (e.g. with mbart-large-50) to force generation of the target language. Are there any pre-defined language codes for mt5, or do I have to add additional special tokens myself ?
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
Facebook mbart multilingual translation | 0 | 479 | February 1, 2023 | |
Can we force first token by model.config.forced_bos_token_id? | 0 | 638 | April 12, 2022 | |
How to train mBart or any multilingual model for translation task | 0 | 249 | January 4, 2023 | |
How can i fine tune a mt5 model cross lingual summarization | 0 | 189 | October 12, 2023 | |
How to constrain mBart decoding to generate English-only output? | 0 | 417 | August 31, 2022 |