This project implements and compares baseline, LoRA, and LoRA + Dynamic Adapter models for sequence-to-sequence learning tasks using the CNN/DailyMail dataset. Below are the instructions to preprocess data, train models, evaluate performance, analyze errors, and launch the Streamlit app for results visualization.

Ensure you have the following installed:

• Python 3.7 or higher
• Required Python packages (install via requirements.txt if available):

  pip install -r requirements.txt

1. Data Preprocessing Preprocess the dataset to prepare it for model training:

python data_preprocessing.py --dataset_name cnn_dailymail --tokenizer_name t5-base

2. Model Training Train the models using the preprocessed dataset:

Baseline Model:

python model_training.py --model_name t5-small --tokenizer_name t5-small

LoRA Model:

python model_training.py --model_name t5-small --tokenizer_name t5-small --use_lora

LoRA + Dynamic Adapter Model:

python model_training.py --model_name t5-small --tokenizer_name t5-small --use_lora --use_adapter

3. Evaluation Evaluate each model on the validation dataset: Baseline Model:

python evaluation.py --model_dir ./saved_model_baseline --tokenizer_name t5-small --output_file evaluation_baseline.txt

LoRA Model:

python evaluation.py --model_dir ./saved_model_lora --tokenizer_name t5-small --output_file evaluation_lora.txt

LoRA + Dynamic Adapter Model:

python evaluation.py --model_dir ./saved_model_lora_adapter --tokenizer_name t5-small --output_file evaluation_lora_adapter.txt

4. Error Analysis Perform error analysis on the LoRA + Dynamic Adapter model:

python error_analysis.py --model_dir ./saved_model_lora_adapter --tokenizer_name t5-small --output_file error_analysis_results.txt

5. Launch Streamlit App Visualize and compare results using the Streamlit app:

streamlit run app.py