- Who it helps
- Developers studying deterministic inference and compression.
- Availability
- Public alpha 0.1.0 code is available under Apache 2.0. Model weights have their own terms and add roughly 127 MB of storage.
Why integer arithmetic matters
The author reports matching results across seven backends. Reproducible predictions let a decoder reconstruct the same bytes without relying on small floating-point differences.
Where it falls short
Reported compression ratios exclude the model download. Throughput is low, especially on CPU, and the README identifies earlier language-model compressors with better speed or compression.
What you can do
- Read the comparison and overhead notes before deciding whether the experiment fits a research task. Open reference ↗
Where the evidence stops
No files were compressed here. Portability and performance remain the author’s test results.
Sources read. Sources checked October 20, 2026 UTC. No external product workflow, model training or inference was executed.
Sources & further reading
GitHub · implementation
TACO integer-only lossless compression ↗
Project-owned implementation or interface showing availability, requirements and practical constraints.
Rohit Kumar · page checked 2026-10-20 · Published 2026-10-19
Coverage note: The source bundle is a bounded sample. Independent replications and matching original discussion across all platforms were not established.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
TACO is an experimental lossless compressor using GPT-2 predictions and integer-only arithmetic to make encoding reproducible across machines.