NanoQuant: Efficient Sub-1-Bit Quantization of Large Language Models

Weight-only quantization has become a standard approach for efficiently serving large language models (LLMs). However, existing methods fail to efficiently compress models to binary (1-bit) levels, as they either require large amounts of data and compute or incur additional storage. In this work, we propose NanoQuant, a post-training quantization (PTQ) method to compress LLMs to both binary and sub-1-bit levels. NanoQuant formulates quantization as a low-rank binary factorization problem, and compresses full-precision weights to low-rank binary matrices and scales. Specifically, it utilizes an efficient alternating direction method of multipliers (ADMM) solver to precisely initialize latent binary matrices and scales, and then tunes the initialized parameters through a block and model reconstruction process. Consequently, NanoQuant establishes a new Pareto frontier in low-memory post-training quantization, and enables sub-1-bit compression. NanoQuant makes large-scale deployment feasible on consumer hardware. For example, it compresses Llama-2-70B by 24$\times$ in just 13 hours on a single H100, enabling a 70B model to operate on a consumer 8 GB GPU.

Large language models are powerful but difficult to run because they require huge amounts of memory. Compressing them to extremely small sizes, such as 1-bit or below, is hard without losing quality or using expensive retraining. NanoQuant addresses this by finding compact binary representations of a trained model’s weights and refining them without full retraining. This allows much stronger compression while preserving model performance. As a result, NanoQuant makes very large models easier to deploy on limited hardware. For example, it compresses Llama-2-70B by 24$\times$ in 13 hours on a single H100, enabling it to run on a consumer 8 GB GPU.