ai4anshu/sentence-compression

Getting Started

Installation

conda environment

conda env create --name NAME --file=environment.yaml

The Project is designed around several scripts that simulate a typical machine learning workflow. Starting with data preparation after preparing data, training model, evaluation and inference model. google/t5-small model was being trained on above dataset for 10 epochs. Later inference ran on evaluation data, performance metrics and evaluation results were stored inside result subdirectory of project directory.

I added Makefile which can be used to run python scripts separately using following bash commands.

make data
make train
make eval
make inference

run is a bash command which can aggregately run entire project.

make run

clean is a bash command which can be used to clean the previous runs.

make clean

Performance metrics stores into performance.json file inside results directory.

{
    "rouge1": 0.79689240266461,
    "rouge2": 0.7606140631154827,
    "rougeL": 0.7733855633904199,
    "rougeLsum": 0.7734703253159519
}

And also, eval_results.csv containing predictions of evaluation file.

original	compressed	predictions
sentence1	compress1	prediction1
sentence2	compress2	prediction2
:	:	:

References:

Note:

Download trained checkpoint from given drive link checkpoint