English

NativeTernary: A Self-Delimiting Binary Encoding with Unary Run-Length Hierarchy Markers for Ternary Neural Network Weights, Structured Data, and General Computing Infrastructure

Machine Learning 2026-04-09 v2 Signal Processing

Abstract

BitNet b1.58 (Ma et al., 2024) demonstrates that large language models can operate entirely on ternary weights {-1, 0, +1}, yet no native binary wire format exists for such models. NativeTernary closes this gap. Benchmarked against GGUF on the real BitNet b1.58 2B4T architecture (24 layers, ~170 tensors, 2B parameters): NativeTernary encodes ternary weights at exactly 2.000 bits per weight -- 1.31x smaller than GGUF Q2_K and 4.0x smaller than GGUF int8 -- while reducing boundary and framing overhead by 460x (91 bytes vs ~42KB of GGUF tensor headers). Encode throughput: 47--69 MB/s. Decode throughput: 35--45 MB/s on commodity hardware. The decoder is a 10-line stateless state machine resilient to bitstream corruption.

Cite

@article{arxiv.2604.03336,
  title  = {NativeTernary: A Self-Delimiting Binary Encoding with Unary Run-Length Hierarchy Markers for Ternary Neural Network Weights, Structured Data, and General Computing Infrastructure},
  author = {Maharshi Savdhariya},
  journal= {arXiv preprint arXiv:2604.03336},
  year   = {2026}
}

Comments

v2: benchmark results added. Real BitNet b1.58 2B4T architecture analysis: NativeTernary framing overhead 460x smaller than GGUF tensor headers (91 bytes vs 42KB). 1.31x smaller than GGUF Q2_K. C implementation: https://github.com/sm45118/nativeternary

R2 v1 2026-07-01T11:53:19.199Z