In this paper, we present BitNet Distillation (BitDistill), a lightweight pipeline that fine-tunes off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (i.e., ternary weights {-1, 0, 1}) for specific downstream tasks, achieving strong task-specific performance with minimal computational cost. Specifically, BitDistill incorporates three key techniques: the SubLN module, as introduced in BitNet; multi-head attention distillation, based on MiniLM; and continual pre-training, which serves as a crucial warm-up step to mitigate the scalability issue of the performance gap between finetuned full-precision and 1.58-bit LLMs on specific tasks. Experimental results show that BitDistill achieves performance comparable to the full-precision counterpart models across model size, while enabling up to 10x memory savings and 2.65x faster inference on CPUs. Code is available at https://github.com/microsoft/BitNet.

Deploying Large Language Models (LLMs) on edge devices remains challenging due to their quadratically increasing computations with the sequence length. Existing studies for dynamic attention pruning are designed for hardware with massively parallel computation capabilities, such as GPUs or TPUs, and aim at long context lengths (e.g., 64K), making them unsuitable for edge scenarios. We present DeltaLLM, a training-free framework that exploits temporal sparsity in attention patterns to enable efficient LLM inference across both the prefilling and decoding stages, on resource-constrained edge devices. DeltaLLM introduces an accuracy- and memory-aware delta matrix construction strategy that introduces temporal sparsity, and a context-aware hybrid attention mechanism that combines full attention in a local context window with delta approximation outside it to increase accuracy. We evaluate our framework on the edge-device-friendly BitNet-b1.58-2B-4T model and Llama3.2-1B-Instruct model across diverse language tasks. The results show that on BitNet, our framework increases the attention sparsity from 0% to 60% during the prefilling stage with slight accuracy improvement on the WG task, and 0% to 57% across both the prefilling and decoding stages, with even higher F1 score from 29.63 to 30.97 on SQuAD-v2 task. On the Llama model, it can also achieve up to 60% sparsity during the prefilling stage and around 57% across both stages with negligible accuracy drop. These results demonstrate that DeltaLLM offers a promising solution for efficient edge deployment, requiring no fine-tuning and seamlessly integrating with existing inference pipelines.

Large language models (LLMs) have wide applications in the field of natural language processing(NLP), such as GPT-4 and Llama. However, with the exponential growth of model parameter sizes, LLMs bring significant resource overheads. Low-bit quantization, as a key technique, reduces memory usage and computational demands by decreasing the bit-width of model parameters, activations, and gradients. Previous quantization methods for LLMs have largely employed Post-Training Quantization (PTQ) and Quantization-Aware Training (QAT). PTQ does not require any retraining of the original model, while QAT involves optimizing precision during training to achieve the best quantization parameters. The BitNet team proposed a radically different approach, where quantization is performed from the start of model training, utilizing low-precision binary weights during the training process. This approach has led to the emergence of many binary quantization techniques for large language models. This paper provides a comprehensive review of these binary quantization techniques. Specifically, we will introduce binary quantization techniques in deep neural networks and further explore their application to LLMs, reviewing their various contributions, implementations, and applications.

The advent of 1-bit large language models (LLMs), led by BitNet b1.58, has spurred interest in ternary LLMs. Despite this, research and practical applications focusing on efficient edge inference for ternary LLMs remain scarce. To bridge this gap, we introduce Bitnet.cpp, an inference system optimized for BitNet b1.58 and ternary LLMs. Given that mixed-precision matrix multiplication (mpGEMM) constitutes the bulk of inference time in ternary LLMs, Bitnet.cpp incorporates a novel mpGEMM library to facilitate sub-2-bits-per-weight, efficient and lossless inference. The library features two core solutions: Ternary Lookup Table (TL), which addresses spatial inefficiencies of previous bit-wise methods, and Int2 with a Scale (I2_S), which ensures lossless edge inference, both enabling high-speed inference. Our experiments show that Bitnet.cpp achieves up to a 6.25x increase in speed over full-precision baselines and up to 2.32x over low-bit baselines, setting new benchmarks in the field. Additionally, we expand TL to element-wise lookup table (ELUT) for low-bit LLMs in the appendix, presenting both theoretical and empirical evidence of its considerable potential. Bitnet.cpp is publicly available at https://github.com/microsoft/BitNet/tree/paper , offering a sophisticated solution for the efficient and practical deployment of edge LLMs.

Although recent quantized Large Language Models (LLMs), such as BitNet, have paved the way for significant reduction in memory usage during deployment with binary or ternary weights, training these models still demands substantial memory footprints. This is partly because high-precision (i.e., unquantized) weight matrices required for straight-through estimation must be maintained throughout the whole training process. To address this, we explore the potential of directly updating the quantized low-precision weight matrices without relying on the straight-through estimator during backpropagation, thereby saving memory usage during training. Specifically, we employ a stochastic rounding technique to minimize information loss caused by the use of low-bit weights throughout training. Experimental results on our LLaMA-structured models indicate that (1) training with only low-precision weights is feasible even when they are constrained to ternary values, (2) extending the bit width to 8 bits results in only a 5% loss degradation compared to BitNet b1.58 while offering the potential for reduced memory usage during training, and (3) our models can also perform inference using ternary weights, showcasing their flexibility in deployment.

Contemporary machine learning models, such as language models, are powerful, but come with immense resource requirements both at training and inference time. It has been shown that decoder-only language models can be trained to a competitive state with ternary weights (1.58 bits per weight), facilitating efficient inference. Here, we start our exploration with non-transformer model architectures, investigating 1.58-bit training for multi-layer perceptrons and graph neural networks. Then, we explore 1.58-bit training in other transformer-based language models, namely encoder-only and encoder-decoder models. Our results show that in all of these settings, 1.58-bit training is on par with or sometimes even better than the standard 32/16-bit models.

Recent advances in 1-bit Large Language Models (LLMs), such as BitNet and BitNet b1.58, present a promising approach to enhancing the efficiency of LLMs in terms of speed and energy consumption. These developments also enable local LLM deployment across a broad range of devices. In this work, we introduce bitnet.cpp, a tailored software stack designed to unlock the full potential of 1-bit LLMs. Specifically, we develop a set of kernels to support fast and lossless inference of ternary BitNet b1.58 LLMs on CPUs. Extensive experiments demonstrate that bitnet.cpp achieves significant speedups, ranging from 2.37x to 6.17x on x86 CPUs and from 1.37x to 5.07x on ARM CPUs, across various model sizes. The code is available at https://github.com/microsoft/BitNet.

The increasing size of large language models has posed challenges for deployment and raised concerns about environmental impact due to high energy consumption. In this work, we introduce BitNet, a scalable and stable 1-bit Transformer architecture designed for large language models. Specifically, we introduce BitLinear as a drop-in replacement of the nn.Linear layer in order to train 1-bit weights from scratch. Experimental results on language modeling show that BitNet achieves competitive performance while substantially reducing memory footprint and energy consumption, compared to state-of-the-art 8-bit quantization methods and FP16 Transformer baselines. Furthermore, BitNet exhibits a scaling law akin to full-precision Transformers, suggesting its potential for effective scaling to even larger language models while maintaining efficiency and performance benefits.

We present a novel regularization scheme for training deep neural networks. The parameters of neural networks are usually unconstrained and have a dynamic range dispersed over the real line. Our key idea is to control the expressive power of the network by dynamically quantizing the range and set of values that the parameters can take. We formulate this idea using a novel end-to-end approach that regularizes the traditional classification loss function. Our regularizer is inspired by the Minimum Description Length principle. For each layer of the network, our approach optimizes a translation and scaling factor along with integer-valued parameters. We empirically compare BitNet to an equivalent unregularized model on the MNIST and CIFAR-10 datasets. We show that BitNet converges faster to a superior quality solution. Additionally, the resulting model is significantly smaller in size due to the use of integer parameters instead of floats.