Ko Trocr Base Nsmc News Chatbot

🚀 TrOCR for Korean Language (PoC)

TrOCR hasn't released a multilingual model including Korean yet. This project trained a Korean model for PoC purpose.

🚀 Quick Start

TrOCR has not launched a multilingual model that includes Korean. Therefore, we trained a Korean model for proof - of - concept (PoC). Based on this model, it is advisable to gather more data for additional first - stage training or conduct second - stage fine - tuning.

✨ Features

• This is a Korean - specific TrOCR model for PoC.
• It uses publicly available datasets for training.
• The code for data collection and model training is open - sourced on GitHub.

📦 Installation

The README does not provide specific installation steps, so this section is skipped.

💻 Usage Examples

Basic Usage

from transformers import TrOCRProcessor, VisionEncoderDecoderModel, AutoTokenizer
import requests 
from io import BytesIO
from PIL import Image

processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten") 
model = VisionEncoderDecoderModel.from_pretrained("daekeun-ml/ko-trocr-base-nsmc-news-chatbot")
tokenizer = AutoTokenizer.from_pretrained("daekeun-ml/ko-trocr-base-nsmc-news-chatbot")

url = "https://raw.githubusercontent.com/aws-samples/sm-kornlp/main/trocr/sample_imgs/news_1.jpg"
response = requests.get(url)
img = Image.open(BytesIO(response.content))

pixel_values = processor(img, return_tensors="pt").pixel_values 
generated_ids = model.generate(pixel_values, max_length=64)
generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] 
print(generated_text)

📚 Documentation

Collecting data

Text data

We created training data by processing three types of datasets:

• News summarization dataset: https://huggingface.co/datasets/daekeun-ml/naver-news-summarization-ko
• Naver Movie Sentiment Classification: https://github.com/e9t/nsmc
• Chatbot dataset: https://github.com/songys/Chatbot_data

For efficient data collection, each sentence was separated by a sentence separator library (Kiwi Python wrapper; https://github.com/bab2min/kiwipiepy), and as a result, 637,401 samples were collected.

Image Data

Image data was generated with TextRecognitionDataGenerator (https://github.com/Belval/TextRecognitionDataGenerator) introduced in the TrOCR paper. Below is a code snippet for generating images.

python3 ./trdg/run.py -i ocr_dataset_poc.txt -w 5 -t {num_cores} -f 64 -l ko -c {num_samples} -na 2 --output_dir {dataset_dir}

Training

Base model

The encoder model used facebook/deit-base-distilled-patch16-384 and the decoder model used klue/roberta-base. It is easier than training by starting weights from microsoft/trocr-base-stage1.

Parameters

We used heuristic parameters without separate hyperparameter tuning.

• learning_rate = 4e - 5
• epochs = 25
• fp16 = True
• max_length = 64

Usage

All the code required for data collection and model training has been published on the author's Github:

• https://github.com/daekeun-ml/sm-kornlp-usecases/tree/main/trocr

🔧 Technical Details

The project uses specific encoder and decoder models for training the Korean TrOCR model. It also uses publicly available datasets for text and generates image data using a specific tool. The training parameters are set heuristically without separate hyperparameter tuning.

📄 License

The project is licensed under the MIT license.