2025ML-project-neural_compression/README.md

neural compression

Running locally

uv sync --all-extras

Example usage:

# Fetching
python main.py --debug train --method fetch \
  --dataset enwik9 --data-root /path/to/datasets

# Training
python main.py --debug train --method optuna \
  --dataset enwik9 --data-root /path/to/datasets \
  --model cnn --model-save-path /path/to/optuna-model
python main.py --debug --results /path/to/results train --method full \
  --dataset enwik9 --data-root /path/to/datasets \
  --model-load-path /path/to/optuna-model --model-save-path /path/to/full-model

# Compressing
python benchmark.py --debug compress \
  --model-load-path /path/to/full-model \
  --input-file inputfile --output-file outputfile

Testing compression:

bash config/download_datasets.sh config/urls.txt /home/tdpeuter/data/ml-inputs
bash config/generate_csv.sh > config/sub.csv
bash config/local.sh

Running on the Ghent University HPC

See the Infrastructure docs for more information about the clusters.

module swap cluster/joltik # Specify the (GPU) cluster, {joltik,accelgor,litleo}

qsub job.pbs               # Submit job
qstat                      # Check status