NNCP: Lossless Data Compression with Neural Networks

NNCP is an experiment to build a practical lossless data compressor with neural networks. The latest version uses a Transformer model.

The algorithms and results are described in nncp_v2.pdf. The previous LSTM based version is described in nncp.pdf.

Compression ratio

Result for enwik8:

Result for enwik9:

Program	Compr. size (bytes)	Ratio (bpb)	Program size (zip, bytes)	Total (bytes)
gzip	322 591 995	2.58	38 801	322 630 796
xz	197 331 816	1.58	36 752	197 368 568
CMIX (v18)	115 714 367	0.926	208 961	115 923 328
NNCP (2021-01-03)	114 217 584	0.914	99 671	114 317 255

* The results for the other programs are from the Large Text Compression Benchmark.

NNCP v2 (Python+PyTorch, GPU required): nncp_v2-2021-01-03.tar.gz (readme.txt)
NNCP v1: Linux version: nncp-2019-11-16.tar.gz. LibNC is currently only provided as object code.
NNCP v1: Precompiled Windows version: nncp-2019-11-16-win64.zip.