This project involves fine-tuning a T5 model to generate automated responses for customer support queries using the dataset from Kaludi/Customer-Support-Responses.
The objective is to build a model capable of generating meaningful and accurate responses to customer queries. The T5 model, pre-trained by Google, has been fine-tuned on a dataset specific to customer support to achieve this task. This project encompasses data preprocessing, model training, evaluation using the BLEU metric, and response generation.
The model is trained using a sequence-to-sequence approach where the input is a customer query and the output is the generated response. A specific prefix "Assure the customer and provide specific help" is added to each query to help the model understand the context and the task it needs to perform.
Experience the model in action through Streamlit app.
Ensure you have the following libraries installed after cloning the repository:
# Create and activate a virtual environment using venv
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
# Alternatively, using conda
conda create --name <env-name> --file requirements.txt
conda activate <env-name>Execute the model_training.ipynb notebook cell by cell to ensure proper functionality.
To adjust the model parameters, edit the hyperparameters.py file and rerun the notebook for training.
Most models trained with 5-10 epochs performed well. However, the model using the latest pretext specified in hyperparameters.py yielded the best results so far.
