mbart ja en

Introduction

MBART-JA-EN is a translation model fine-tuned from Facebook's mBART-large-cc25. It is specifically designed for Japanese to English translation using the JESC dataset.

Architecture

The model is based on the mBART (Multilingual BART) architecture, which supports text-to-text generation tasks. It utilizes a transformer architecture and is implemented in PyTorch, allowing it to handle multilingual tasks effectively.

Training

The model was fine-tuned using the JESC dataset, a large collection of Japanese-English sentence pairs. The tokenizer employed is based on SentencePiece, trained specifically on the JESC dataset.

Guide: Running Locally

To run the MBART-JA-EN model locally, follow these steps:

1. Install Transformers Library: Ensure you have the Hugging Face Transformers library installed.

   pip install transformers
   

2. Load the Model and Tokenizer:

   from transformers import MBartForConditionalGeneration, MBartTokenizer
   
   tokenizer = MBartTokenizer.from_pretrained("ken11/mbart-ja-en")
   model = MBartForConditionalGeneration.from_pretrained("ken11/mbart-ja-en")
   

3. Input and Generate Translation:

   inputs = tokenizer("こんにちは", return_tensors="pt")
   translated_tokens = model.generate(**inputs, decoder_start_token_id=tokenizer.lang_code_to_id["en_XX"], early_stopping=True, max_length=48)
   pred = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
   print(pred)
   

For optimal performance, especially with larger datasets, consider using cloud-based solutions like AWS EC2 with GPU instances or Google Cloud Platform's GPUs.

License

The model is distributed under the MIT License, allowing for flexible use and distribution.