Multitask bilingual ticket classification using XLM-RoBERTa

A pipeline for fine-tuning multilingual transformer models on bilingual customer support tickets. Built with 🤗 Transformers, 🧠 Weights & Biases, and 🔥 PyTorch.

Fine-tuned on the ale-dp/bilingual-ticket-classification dataset using the multilingual model FacebookAI/xlm-roberta-base.

⚡ Quickstart

Clone the Repository

git clone https://github.com/alikhalajii/bilingual_ticket_classifier.git
cd bilingual_ticket_classifier

If you want to skip training and directly run the demo using the fine-tuned model:

git clone https://github.com/alikhalajii/bilingual_ticket_classifier.git && cd bilingual_ticket_classifier && git lfs install && git lfs pull

Install the repository as a Python package

pip install -e .

Set up environment variables

Make sure to add your Weights & Biases API key and Hugging Face token to the .env file in the project root: Train the model

python src/bilingual_ticket_classifier/training/train.py

Evaluate the Model

python src/bilingual_ticket_classifier/evaluation/evaluate.py

Launch the Gradio Demo

python demo/app.py

📊 Results

All training runs are logged to Weights & Biases for full transparency and reproducibility.

• Latest W&B Run Dashboard
  Explore training curves, evaluation metrics, and system logs.

• Summary JSON Snapshot
  Quick access to final metrics.