[2603.27914] ITQ3_S: High-Fidelity 3-bit LLM Inference via Interleaved Ternary Quantization with Rotation-Domain Smoothing

Computer Science > Machine Learning arXiv:2603.27914 (cs) [Submitted on 30 Mar 2026] Title:ITQ3_S: High-Fidelity 3-bit LLM Inference via Interleaved Ternary Quantization with Rotation-Domain Smoothing Authors:Edward J. Yoon View a PDF of the paper titled ITQ3_S: High-Fidelity 3-bit LLM Inference via Interleaved Ternary Quantization with Rotation-Domain Smoothing, by Edward J. Yoon View PDF HTML (experimental) Abstract:We present \textbf{ITQ3\_S} (Interleaved Ternary Quantization -- Specialized), a novel 3-bit weight quantization format for large language models (LLMs) that integrates \textbf{TurboQuant (TQ)}, a rotation-domain adaptive quantization strategy based on the Fast Walsh-Hadamard Transform (FWHT). Conventional 3-bit quantization methods suffer from catastrophic precision loss caused by heavy-tailed weight distributions and inter-channel outliers. ITQ3\_S addresses this fundamental limitation by pre-rotating the weight space via FWHT prior to quantization, effectively spreading outlier energy across the entire vector and inducing a near-Gaussian distribution amenable to uniform ternary coding. Critically, we derive a mathematically rigorous dequantization procedure that inverts the FWHT exactly using a 256-point Inverse Walsh-Hadamard Transform fused into the CUDA shared-memory loading stage, ensuring zero-error round-trip fidelity between offline quantization and online inference. We prove that for any weight vector $\mathbf{w} \in \mathbb{R}^{256}$ processed by ...