Evaluating small neural networks for general-purpose lossy data compression
This repository has been archived on 2025-12-23. You can view files and clone it, but you cannot make any changes to it's state, such as pushing and creating new issues, pull requests or comments.
Find a file
2025-12-11 20:36:24 +01:00
results
src fix: encoding binary now SUPAH fast 2025-12-11 20:36:24 +01:00
.gitignore
.python-version
benchmark.py
job.pbs chore: Job cnn-enwik9-optuna 2025-12-10 10:34:06 +01:00
main.py fixup! feat: Size arg 2025-12-10 16:34:53 +01:00
pyproject.toml feat (WIP): Compress 2025-12-10 21:13:09 +01:00
README.md
uv.lock feat (WIP): Compress 2025-12-10 21:13:09 +01:00

neural compression

Example usage:

python main.py --debug train --dataset enwik9 --data-root ~/data/datasets/ml --method optuna --model transformer --model-save-path ~/data/ml-models/test-transformer.pt

python benchmark.py --debug train --dataset enwik9 --data-root ~/data/datasets/ml --method optuna --model cnn --model-save-path ~/data/ml-models/test-cnn.pt

Running locally

uv sync --all-extras

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