NativeTernary: 一种带有 unary 运行长度层级标记的自限二进制编码，用于三进制神经网络权重、结构化数据和通用计算基础设施
NativeTernary: A Self-Delimiting Binary Encoding with Unary Run-Length Hierarchy Markers for Ternary Neural Network Weights, Structured Data, and General Computing Infrastructure
AI总结：
NativeTernary 提出了一种自限二进制编码，用于三进制神经网络权重和结构化数据，相比 GGUF 在存储效率和性能上有显著提升。
摘要：
BitNet b1.58 (Ma et al., 2024) 证明了大型语言模型可以完全在三进制权重 {-1, 0, +1} 上运行，但没有原生的二进制线格式用于此类模型。NativeTernary 填补了这一空白。在真实 BitNet b1.58 2B4T 架构 (24 层，~170 个张量，2B 参数) 上与 GGUF 相比：NativeTernary 对三进制权重的编码精确为 2.000 bits per weight -- 1.31x 小于 GGUF Q2_K，4.0x 小于 GGUF int8，同时减少边界和框架开销 460x (91 字节 vs ~42KB 的 GGUF 张量标题)。编码吞吐量：47--69 MB/s。解码吞吐量：在商用硬件上为 35--45 MB/s。解码器是一个 10 行无状态状态机，对位流损坏具有鲁棒性。
英文摘要：
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.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
