Mixture of Scales: Memory-Efficient Token-Adaptive Binarization for Large Language Models
This addresses the memory efficiency vs. performance trade-off in deploying large language models, offering an incremental improvement over existing binarization methods.
The paper tackles the problem of binarization reducing the linguistic effectiveness of large language models by introducing Mixture of Scales (BinaryMoS), a token-adaptive binarization technique that dynamically merges scaling experts per token, resulting in performance surpassing conventional binarization and even 2-bit quantization methods while maintaining similar model size.
Binarization, which converts weight parameters to binary values, has emerged as an effective strategy to reduce the size of large language models (LLMs). However, typical binarization techniques significantly diminish linguistic effectiveness of LLMs. To address this issue, we introduce a novel binarization technique called Mixture of Scales (BinaryMoS). Unlike conventional methods, BinaryMoS employs multiple scaling experts for binary weights, dynamically merging these experts for each token to adaptively generate scaling factors. This token-adaptive approach boosts the representational power of binarized LLMs by enabling contextual adjustments to the values of binary weights. Moreover, because this adaptive process only involves the scaling factors rather than the entire weight matrix, BinaryMoS maintains compression efficiency similar to traditional static binarization methods. Our experimental results reveal that BinaryMoS surpasses conventional binarization techniques in various natural language processing tasks and even outperforms 2-bit quantization methods, all while maintaining similar model size to static binarization techniques.