Nacrith — Lossless text compression via ensemble neural arithmetic coding. Combines SmolLM2-135M language model with context mixing, adaptive prediction, and high-precision CDF coding. 3.1x better than gzip, and outperforming CMIX, ts_zip, LLMZip and FineZip. Fully lossless. GPU accelerated.
World's best JSON/NDJSON compressor. Format-aware lossless compression engine in Rust - schema inference, columnar reorg, type-specific encoding. 2-3x better than zstd on structured data.
An implementation of fpaq0p in Zig (and something more...)
A lossless context-mixing compressor and archiver in a single 4,500-line C file. Beats zpaq -m5 on all twelve Silesia files; its -9 total would place just inside the top 50 of 320 entries on Mahoney's Silesia benchmark. At -5 it is simultaneously 2.2% smaller, 1.6x faster and 43% lighter on memory than zpaq -m5.
High-ratio specialized compressor built in a 48h Human+AI experiment. Uses context mixing to beat zpaq/xz on structured binaries & signals.
Dense Zip is an archiver tuned for the smallest possible output — recompression, context mixing, and backend racing. Beats gzip/xz/zstd/7z on size.
general purpose lossless compression for no one, built by an agent team
Minimal Zip Concept (MZC) - Learning-oriented lossless compression engine in Rust featuring RLE, Dictionary, LZ77, BCJ, Delta, and Dynamic Huffman Coding with a real-time GUI plot dashboard.
PawCompress: High-performance hybrid lossless compression engine (LZ77 + Range Coder + Context Mixing)
A lossless DNA compressor in C — context mixing, tolerant match models, reference mode
TokenTrie: Real-time sequence prediction engine built on Dynamic Suffix Tries. Implements Context Mixing and Entropy Scaling to capture variable-order dependencies without neural networks.
To associate your repository with the context-mixing topic, visit your repo's landing page and select "manage topics."