--- library_name: transformers language: - en - tpi base_model: - Helsinki-NLP/opus-mt-en-tpi pipeline_tag: translation datasets: - RickBrannan/tpi_eng_sentence_pairs license: apache-2.0 --- # Model Card for Model ID This model is a fine-tune of the `opus-mt-en-tpi` model. ## Model Details ### Model Description - **Model type:** Translation - **Language(s) (NLP):** English, Tok Pisin - **Finetuned from model [optional]:** https://huggingface.co/Helsinki-NLP/opus-mt-en-tpi ### Model Sources - **Repository:** https://huggingface.co/Helsinki-NLP/opus-mt-en-tpi ## Uses This model is intended for translation of English material into Tok Pisin. It is fine-tuned on material from Bible stories, from selected articles of a Bible Dictionary translated into Tok Pisin, and from translation of the deuterocanon (apocrypha) into Tok Pisin. ## How to Get Started with the Model Use the code below to get started with the model. ``` from transformers import pipeline pipe = pipeline("translation", model="RickBrannan/opus-mt-en-tpi-finetune") translation = pipe(">>tpi<< In the beginning, God created the heavens and the earth.") ``` ## Training Details ### Training Data * 1,100+ English-Tok Pisin sentence pairs based on "Open Bible Stories" from unfoldingWord: https://git.door43.org/door43-Catalog/tpi_obs * 2,600+ English-Tok Pisin sentence pairs based on translations from English into Tok Pisin pulled from the Bible Aquifer: https://aquifer.bible * 4,150+ English-Tok Pisin sentence pairs based on translation of deuterocanonical books of the Bible into Tok Pisin: https://ebible.org/Scriptures/details.php?id=tpi The first two sources are available via CC-BY-SA license and available in the [RickBrannan/tpi_eng_sentence_pairs](https://huggingface.co/datasets/RickBrannan/tpi_eng_sentence_pairs) dataset. The Tok Pisin of the last source is available via CC-BY-NC-ND from the website listed. For the English, we used the text of the deuterocanon of the [World English Bible (WEB)](https://ebible.org/Scriptures/details.php?id=eng-web) where references matched. Where WEB did not have a reference, we used the [deuterocanon of the English Revised Version (RV)](https://ebible.org/Scriptures/details.php?id=eng-rv). #### Testing Data Testing data was 10% of the sentences from the above specified training data.