minhtoan's picture
Update README.md
4c0c1bd verified
metadata
language:
  - vi
  - lo
tags:
  - translation
license: mit
widget:
  - text: Tôi muốn mua một cuốn sách
inference:
  parameters:
    max_length: 200
pipeline_tag: translation
library_name: transformers

Vietnamese to Lao Translation Model

In the domain of natural language processing (NLP), the development of translation models tailored for low-resource languages represents a critical endeavor to facilitate cross-cultural communication and knowledge exchange. In response to this challenge, we present a novel and impactful contribution: a translation model specifically designed to bridge the linguistic gap between Lao and Vietnamese.

Lao, a language spoken primarily in Laos and parts of Thailand, presents inherent challenges for machine translation due to its low-resource nature, characterized by limited parallel corpora and linguistic resources. Vietnamese, a language spoken by millions worldwide, shares some linguistic similarities with Lao, making it an ideal target language for translation purposes.

Leveraging the power of the Transformer-based T5 model, we have developed a robust translation system for the Vietnamese-Lao language pair. The T5 model, renowned for its versatility and effectiveness across various NLP tasks, serves as the cornerstone of our approach. Through fine-tuning on a curated dataset of Lao-Vietnamese parallel texts, we have endeavored to enhance translation accuracy and fluency, thus enabling smoother communication between speakers of these languages.

Our work represents a significant advancement in the field of machine translation, particularly for low-resource languages like Lao. By harnessing state-of-the-art NLP techniques and focusing on the specific linguistic nuances of the Lao-Vietnamese language pair, we aim to provide a valuable resource for facilitating cross-linguistic communication and cultural exchange.

How to use

On GPU

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("minhtoan/t5-translate-vietnamese-lao")  
model = AutoModelForSeq2SeqLM.from_pretrained("minhtoan/t5-translate-vietnamese-lao")
model.cuda()
src = "Tôi muốn mua một cuốn sách"
tokenized_text = tokenizer.encode(src, return_tensors="pt").cuda()
model.eval()
translate_ids = model.generate(tokenized_text, max_length=200)
output = tokenizer.decode(translate_ids[0], skip_special_tokens=True)
output

'ຂ້ອຍຢາກຊື້ປຶ້ມ'

On CPU

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("minhtoan/t5-translate-vietnamese-lao")  
model = AutoModelForSeq2SeqLM.from_pretrained("minhtoan/t5-translate-vietnamese-lao")
src = "Tôi muốn mua một cuốn sách"
input_ids = tokenizer(src, max_length=200, return_tensors="pt", padding="max_length", truncation=True).input_ids
outputs = model.generate(input_ids=input_ids, max_new_tokens=200)
output = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
output

'ຂ້ອຍຢາກຊື້ປຶ້ມ'

Author

Phan Minh Toan