dhrumeen/small_summarization_model

Summarization with FLAN-T5

This project demonstrates a workflow to fine-tune the FLAN-T5 model for dialogue summarization using the DialogSum dataset. It includes data preprocessing, tokenization, training, and evaluation of the model.

Installation

1. Create a virtual environment

python3 -m venv .venv
source .venv/bin/activate

2. Run the below command

pip install -r requirements.txt

Model Features:

1. Dataset: Uses the DialogSum dataset for summarization tasks.
2. Model: Fine-tunes the FLAN-T5 model (google/flan-t5-base) for generating conversation summaries.
3. GPU Support: Automatically detects and utilizes CUDA or MPS if available for faster computation.
4. Tokenization: Prepares data for training using the transformers library.
5. Training: Implements a training pipeline with Trainer for fine-tuning the model.
6. Evaluation: Compares summaries generated by the baseline, the fine-tuned model, and human-written summaries.

Customization

1. Training Configuration: Modify TrainingArguments to customize learning rate, number of epochs, and other parameters.
2. Dataset: Replace the DialogSum dataset with your dataset for different summarization tasks.