Markus Nagel

Markus Nagel, Marios Fournarakis, Yelysei Bondarenko et al.

When training neural networks with simulated quantization, we observe that quantized weights can, rather unexpectedly, oscillate between two grid-points. The importance of this effect and its impact on quantization-aware training (QAT) are not well-understood or investigated in literature. In this paper, we delve deeper into the phenomenon of weight oscillations and show that it can lead to a significant accuracy degradation due to wrongly estimated batch-normalization statistics during inference and increased noise during training. These effects are particularly pronounced in low-bit ($\leq$ 4-bits) quantization of efficient networks with depth-wise separable layers, such as MobileNets and EfficientNets. In our analysis we investigate several previously proposed QAT algorithms and show that most of these are unable to overcome oscillations. Finally, we propose two novel QAT algorithms to overcome oscillations during training: oscillation dampening and iterative weight freezing. We demonstrate that our algorithms achieve state-of-the-art accuracy for low-bit (3 & 4 bits) weight and activation quantization of efficient architectures, such as MobileNetV2, MobileNetV3, and EfficentNet-lite on ImageNet. Our source code is available at {https://github.com/qualcomm-ai-research/oscillations-qat}.

Mart van Baalen, Andrey Kuzmin, Suparna S Nair et al.

Recently, the idea of using FP8 as a number format for neural network training has been floating around the deep learning world. Given that most training is currently conducted with entire networks in FP32, or sometimes FP16 with mixed-precision, the step to having some parts of a network run in FP8 with 8-bit weights is an appealing potential speed-up for the generally costly and time-intensive training procedures in deep learning. A natural question arises regarding what this development means for efficient inference on edge devices. In the efficient inference device world, workloads are frequently executed in INT8. Sometimes going even as low as INT4 when efficiency calls for it. In this whitepaper, we compare the performance for both the FP8 and INT formats for efficient on-device inference. We theoretically show the difference between the INT and FP formats for neural networks and present a plethora of post-training quantization and quantization-aware-training results to show how this theory translates to practice. We also provide a hardware analysis showing that the FP formats are somewhere between 50-180% less efficient in terms of compute in dedicated hardware than the INT format. Based on our research and a read of the research field, we conclude that although the proposed FP8 format could be good for training, the results for inference do not warrant a dedicated implementation of FP8 in favor of INT8 for efficient inference. We show that our results are mostly consistent with previous findings but that important comparisons between the formats have thus far been lacking. Finally, we discuss what happens when FP8-trained networks are converted to INT8 and conclude with a brief discussion on the most efficient way for on-device deployment and an extensive suite of INT8 results for many models.

Yelysei Bondarenko, Markus Nagel, Tijmen Blankevoort

Transformer models have been widely adopted in various domains over the last years, and especially large language models have advanced the field of AI significantly. Due to their size, the capability of these networks has increased tremendously, but this has come at the cost of a significant increase in necessary compute. Quantization is one of the most effective ways to reduce the computational time and memory consumption of neural networks. Many studies have shown, however, that modern transformer models tend to learn strong outliers in their activations, making them difficult to quantize. To retain acceptable performance, the existence of these outliers requires activations to be in higher bitwidth or the use of different numeric formats, extra fine-tuning, or other workarounds. We show that strong outliers are related to very specific behavior of attention heads that try to learn a "no-op" or just a partial update of the residual. To achieve the exact zeros needed in the attention matrix for a no-update, the input to the softmax is pushed to be larger and larger during training, causing outliers in other parts of the network. Based on these observations, we propose two simple (independent) modifications to the attention mechanism - clipped softmax and gated attention. We empirically show that models pre-trained using our methods learn significantly smaller outliers while maintaining and sometimes even improving the floating-point task performance. This enables us to quantize transformers to full INT8 quantization of the activations without any additional effort. We demonstrate the effectiveness of our methods on both language models (BERT, OPT) and vision transformers.

Andrey Kuzmin, Mart Van Baalen, Yuwei Ren et al.

When quantizing neural networks for efficient inference, low-bit integers are the go-to format for efficiency. However, low-bit floating point numbers have an extra degree of freedom, assigning some bits to work on an exponential scale instead. This paper in-depth investigates this benefit of the floating point format for neural network inference. We detail the choices that can be made for the FP8 format, including the important choice of the number of bits for the mantissa and exponent, and show analytically in which settings these choices give better performance. Then we show how these findings translate to real networks, provide an efficient implementation for FP8 simulation, and a new algorithm that enables the learning of both the scale parameters and the number of exponent bits in the FP8 format. Our chief conclusion is that when doing post-training quantization for a wide range of networks, the FP8 format is better than INT8 in terms of accuracy, and the choice of the number of exponent bits is driven by the severity of outliers in the network. We also conduct experiments with quantization-aware training where the difference in formats disappears as the network is trained to reduce the effect of outliers.

Andrey Kuzmin, Markus Nagel, Mart van Baalen et al.

Neural network pruning and quantization techniques are almost as old as neural networks themselves. However, to date only ad-hoc comparisons between the two have been published. In this paper, we set out to answer the question on which is better: neural network quantization or pruning? By answering this question, we hope to inform design decisions made on neural network hardware going forward. We provide an extensive comparison between the two techniques for compressing deep neural networks. First, we give an analytical comparison of expected quantization and pruning error for general data distributions. Then, we provide lower bounds for the per-layer pruning and quantization error in trained networks, and compare these to empirical error after optimization. Finally, we provide an extensive experimental comparison for training 8 large-scale models on 3 tasks. Our results show that in most cases quantization outperforms pruning. Only in some scenarios with very high compression ratio, pruning might be beneficial from an accuracy standpoint.

Nilesh Prasad Pandey, Markus Nagel, Mart van Baalen et al.

Neural network quantization is frequently used to optimize model size, latency and power consumption for on-device deployment of neural networks. In many cases, a target bit-width is set for an entire network, meaning every layer get quantized to the same number of bits. However, for many networks some layers are significantly more robust to quantization noise than others, leaving an important axis of improvement unused. As many hardware solutions provide multiple different bit-width settings, mixed-precision quantization has emerged as a promising solution to find a better performance-efficiency trade-off than homogeneous quantization. However, most existing mixed precision algorithms are rather difficult to use for practitioners as they require access to the training data, have many hyper-parameters to tune or even depend on end-to-end retraining of the entire model. In this work, we present a simple post-training mixed precision algorithm that only requires a small unlabeled calibration dataset to automatically select suitable bit-widths for each layer for desirable on-device performance. Our algorithm requires no hyper-parameter tuning, is robust to data variation and takes into account practical hardware deployment constraints making it a great candidate for practical use. We experimentally validate our proposed method on several computer vision tasks, natural language processing tasks and many different networks, and show that we can find mixed precision networks that provide a better trade-off between accuracy and efficiency than their homogeneous bit-width equivalents.

Kartik Gupta, Marios Fournarakis, Matthias Reisser et al.

Federated Learning (FL) is a machine learning paradigm to distributively learn machine learning models from decentralized data that remains on-device. Despite the success of standard Federated optimization methods, such as Federated Averaging (FedAvg) in FL, the energy demands and hardware induced constraints for on-device learning have not been considered sufficiently in the literature. Specifically, an essential demand for on-device learning is to enable trained models to be quantized to various bit-widths based on the energy needs and heterogeneous hardware designs across the federation. In this work, we introduce multiple variants of federated averaging algorithm that train neural networks robust to quantization. Such networks can be quantized to various bit-widths with only limited reduction in full precision model accuracy. We perform extensive experiments on standard FL benchmarks to evaluate our proposed FedAvg variants for quantization robustness and provide a convergence analysis for our Quantization-Aware variants in FL. Our results demonstrate that integrating quantization robustness results in FL models that are significantly more robust to different bit-widths during quantized on-device inference.

Nilesh Prasad Pandey, Marios Fournarakis, Chirag Patel et al.

Post-training quantization (PTQ) is the go-to compression technique for large generative models, such as stable diffusion or large language models. PTQ methods commonly keep the softmax activation in higher precision as it has been shown to be very sensitive to quantization noise. However, this can lead to a significant runtime and power overhead during inference on resource-constraint edge devices. In this work, we investigate the source of the softmax sensitivity to quantization and show that the quantization operation leads to a large bias in the softmax output, causing accuracy degradation. To overcome this issue, we propose an offline bias correction technique that improves the quantizability of softmax without additional compute during deployment, as it can be readily absorbed into the quantization parameters. We demonstrate the effectiveness of our method on stable diffusion v1.5 and 125M-size OPT language model, achieving significant accuracy improvement for 8-bit quantized softmax.

Ties van Rozendaal, Tushar Singhal, Hoang Le et al.

Neural video codecs have recently become competitive with standard codecs such as HEVC in the low-delay setting. However, most neural codecs are large floating-point networks that use pixel-dense warping operations for temporal modeling, making them too computationally expensive for deployment on mobile devices. Recent work has demonstrated that running a neural decoder in real time on mobile is feasible, but shows this only for 720p RGB video. This work presents the first neural video codec that decodes 1080p YUV420 video in real time on a mobile device. Our codec relies on two major contributions. First, we design an efficient codec that uses a block-based motion compensation algorithm available on the warping core of the mobile accelerator, and we show how to quantize this model to integer precision. Second, we implement a fast decoder pipeline that concurrently runs neural network components on the neural signal processor, parallel entropy coding on the mobile GPU, and warping on the warping core. Our codec outperforms the previous on-device codec by a large margin with up to 48% BD-rate savings, while reducing the MAC count on the receiver side by $10 \times$. We perform a careful ablation to demonstrate the effect of the introduced motion compensation scheme, and ablate the effect of model quantization.

Jorn Peters, Marios Fournarakis, Markus Nagel et al.

Quantizing neural networks is one of the most effective methods for achieving efficient inference on mobile and embedded devices. In particular, mixed precision quantized (MPQ) networks, whose layers can be quantized to different bitwidths, achieve better task performance for the same resource constraint compared to networks with homogeneous bitwidths. However, finding the optimal bitwidth allocation is a challenging problem as the search space grows exponentially with the number of layers in the network. In this paper, we propose QBitOpt, a novel algorithm for updating bitwidths during quantization-aware training (QAT). We formulate the bitwidth allocation problem as a constraint optimization problem. By combining fast-to-compute sensitivities with efficient solvers during QAT, QBitOpt can produce mixed-precision networks with high task performance guaranteed to satisfy strict resource constraints. This contrasts with existing mixed-precision methods that learn bitwidths using gradients and cannot provide such guarantees. We evaluate QBitOpt on ImageNet and confirm that we outperform existing fixed and mixed-precision methods under average bitwidth constraints commonly found in the literature.

Minseop Park, Jaeseong You, Markus Nagel et al.

Transformer language models such as GPT-2 are difficult to quantize because of outliers in activations leading to a large quantization error. To adapt to the error, one must use quantization-aware training, which entails a fine-tuning process based on the dataset and the training pipeline identical to those for the original model. Pretrained language models, however, often do not grant access to their datasets and training pipelines, forcing us to rely on arbitrary ones for fine-tuning. In that case, it is observed that quantization-aware training overfits the model to the fine-tuning data. For quantization without overfitting, we introduce a quantization adapter (Quadapter), a small set of parameters that are learned to make activations quantization-friendly by scaling them channel-wise. It keeps the model parameters unchanged. By applying our method to the challenging task of quantizing GPT-2, we demonstrate that it effectively prevents the overfitting and improves the quantization performance.

Davide Abati, Haitam Ben Yahia, Markus Nagel et al.

This paper accelerates video perception, such as semantic segmentation and human pose estimation, by levering cross-frame redundancies. Unlike the existing approaches, which avoid redundant computations by warping the past features using optical-flow or by performing sparse convolutions on frame differences, we approach the problem from a new perspective: low-bit quantization. We observe that residuals, as the difference in network activations between two neighboring frames, exhibit properties that make them highly quantizable. Based on this observation, we propose a novel quantization scheme for video networks coined as Residual Quantization. ResQ extends the standard, frame-by-frame, quantization scheme by incorporating temporal dependencies that lead to better performance in terms of accuracy vs. bit-width. Furthermore, we extend our model to dynamically adjust the bit-width proportional to the amount of changes in the video. We demonstrate the superiority of our model, against the standard quantization and existing efficient video perception models, using various architectures on semantic segmentation and human pose estimation benchmarks.

Andrey Kuzmin, Mart van Baalen, Markus Nagel et al.

In this paper, we introduce a novel method of neural network weight compression. In our method, we store weight tensors as sparse, quantized matrix factors, whose product is computed on the fly during inference to generate the target model's weights. We use projected gradient descent methods to find quantized and sparse factorization of the weight tensors. We show that this approach can be seen as a unification of weight SVD, vector quantization, and sparse PCA. Combined with end-to-end fine-tuning our method exceeds or is on par with previous state-of-the-art methods in terms of the trade-off between accuracy and model size. Our method is applicable to both moderate compression regimes, unlike vector quantization, and extreme compression regimes.

Yelysei Bondarenko, Thomas Hehn, Rob Hesselink et al.

Large language models (LLMs) with chain-of-thought reasoning achieve state-of-the-art performance across complex problem-solving tasks, but their verbose reasoning traces and large context requirements make them impractical for edge deployment. These challenges include high token generation costs, large KV-cache footprints, and inefficiencies when distilling reasoning capabilities into smaller models for mobile devices. Existing approaches often rely on distilling reasoning traces from larger models into smaller models, which are verbose and stylistically redundant, undesirable for on-device inference. In this work, we propose a lightweight approach to enable reasoning in small LLMs using LoRA adapters combined with supervised fine-tuning. We further introduce budget forcing via reinforcement learning on these adapters, significantly reducing response length with minimal accuracy loss. To address memory-bound decoding, we exploit parallel test-time scaling, improving accuracy at minor latency increase. Finally, we present a dynamic adapter-switching mechanism that activates reasoning only when needed and a KV-cache sharing strategy during prompt encoding, reducing time-to-first-token for on-device inference. Experiments on Qwen2.5-7B demonstrate that our method achieves efficient, accurate reasoning under strict resource constraints, making LLM reasoning practical for mobile scenarios. Videos demonstrating our solution running on mobile devices are available on our project page.

Kartikeya Bhardwaj, Nilesh Prasad Pandey, Sweta Priyadarshi et al.

In this paper, we propose Sparse High Rank Adapters (SHiRA) that directly finetune 1-2% of the base model weights while leaving others unchanged, thus, resulting in a highly sparse adapter. This high sparsity incurs no inference overhead, enables rapid switching directly in the fused mode, and significantly reduces concept-loss during multi-adapter fusion. Our extensive experiments on LVMs and LLMs demonstrate that finetuning merely 1-2% parameters in the base model is sufficient for many adapter tasks and significantly outperforms Low Rank Adaptation (LoRA). We also show that SHiRA is orthogonal to advanced LoRA methods such as DoRA and can be easily combined with existing techniques.

Marco Federici, Davide Belli, Mart van Baalen et al.

While mobile devices provide ever more compute power, improvements in DRAM bandwidth are much slower. This is unfortunate for large language model (LLM) token generation, which is heavily memory-bound. Previous work has proposed to leverage natural dynamic activation sparsity in ReLU-activated LLMs to reduce effective DRAM bandwidth per token. However, more recent LLMs use SwiGLU instead of ReLU, which results in little inherent sparsity. While SwiGLU activations can be pruned based on magnitude, the resulting sparsity patterns are difficult to predict, rendering previous approaches ineffective. To circumvent this issue, our work introduces Dynamic Input Pruning (DIP): a predictor-free dynamic sparsification approach, which preserves accuracy with minimal fine-tuning. DIP can further use lightweight LoRA adapters to regain some performance lost during sparsification. Lastly, we describe a novel cache-aware masking strategy, which considers the cache state and activation magnitude to further increase cache hit rate, improving LLM token rate on mobile devices. DIP outperforms other methods in terms of accuracy, memory and throughput trade-offs across simulated hardware settings. On Phi-3-Medium, DIP achieves a 46\% reduction in memory and 40\% increase in throughput with $<$ 0.1 loss in perplexity when compared to streaming the dense model from Flash. The open source code for HW simulator, methods, and experiments in this paper is available at https://github.com/Qualcomm-AI-research/dynamic-sparsity .

Tycho F. A. van der Ouderaa, Mart van Baalen, Paul Whatmough et al.

Scalar quantization of large language models (LLMs) is fundamentally limited by information-theoretic bounds. While vector quantization (VQ) overcomes these limits by encoding blocks of parameters jointly, practical implementations must avoid the need for expensive lookup mechanisms or other explicit codebook storage. Lattice approaches address this through highly structured and dense packing. This paper explores the Leech lattice, which, with its optimal sphere packing and kissing configurations at 24 dimensions, is the highest dimensional lattice known with such optimal properties. To make the Leech lattice usable for LLM quantization, we extend an existing search algorithm based on the extended Golay code construction, to i) support indexing, enabling conversion to and from bitstrings without materializing the codebook, ii) allow angular search over union of Leech lattice shells, iii) propose fully-parallelisable dequantization kernel. Together this yields a practical algorithm, namely Leech Lattice Vector Quantization (LLVQ). LLVQ delivers state-of-the-art LLM quantization performance, outperforming recent methods such as Quip\#, QTIP, and PVQ. These results highlight the importance of high-dimensional lattices for scalable, theoretically grounded model compression.

Marco Federici, Boris van Breugel, Paul Whatmough et al.

Quantization can drastically increase the efficiency of large language and vision models, but typically incurs an accuracy drop. Recently, function-preserving transforms (e.g. rotations, Hadamard transform, channel-wise scaling) have been successfully applied to reduce post-training quantization error, yet a principled explanation remains elusive. We analyze linear-layer quantization via the signal-to-quantization-noise ratio (SQNR), showing that for uniform integer quantization at a fixed bit width, SQNR decomposes into (i) the concentration of weights and activations (capturing spread and outliers), and (ii) the alignment of their dominant variation directions. This reveals an actionable insight: beyond concentration - the focus of most prior transforms (e.g. rotations or Hadamard) - improving alignment between weight and activation can further reduce quantization error. Motivated by this, we introduce block Concentration-Alignment Transforms (CAT), a lightweight linear transformation that uses a covariance estimate from a small calibration set to jointly improve concentration and alignment, approximately maximizing SQNR. Experiments across several LLMs show that CAT consistently matches or outperforms prior transform-based quantization methods at 4-bit precision, confirming the insights gained in our framework.

Marco Federici, Riccardo Del Chiaro, Boris van Breugel et al.

Quantization is the key method for reducing inference latency, power and memory footprint of generative AI models. However, accuracy often degrades sharply when activations are quantized below eight bits. Recent work suggests that invertible linear transformations (e.g. rotations) can aid quantization, by reparameterizing feature channels and weights. In this paper, we propose \textit{Sequence Transformation and Mixed Precision} (STaMP) quantization, a novel strategy that applies linear transformations along the \textit{sequence} dimension to exploit the strong local correlation in language and visual data. By keeping a small number of tokens in each intermediate activation at higher precision, we can maintain model accuracy at lower (average) activations bit-widths. We evaluate STaMP on recent LVM and LLM architectures, demonstrating that it significantly improves low bit width activation quantization and complements established activation and weight quantization methods including recent feature transformations.

Yelysei Bondarenko, Riccardo Del Chiaro, Markus Nagel

Large language models (LLMs) are omnipresent, however their practical deployment is challenging due to their ever increasing computational and memory demands. Quantization is one of the most effective ways to make them more compute and memory efficient. Quantization-aware training (QAT) methods, generally produce the best quantized performance, however it comes at the cost of potentially long training time and excessive memory usage, making it impractical when applying for LLMs. Inspired by parameter-efficient fine-tuning (PEFT) and low-rank adaptation (LoRA) literature, we propose LR-QAT -- a lightweight and memory-efficient QAT algorithm for LLMs. LR-QAT employs several components to save memory without sacrificing predictive performance: (a) low-rank auxiliary weights that are aware of the quantization grid; (b) a downcasting operator using fixed-point or double-packed integers and (c) checkpointing. Unlike most related work, our method (i) is inference-efficient, leading to no additional overhead compared to traditional PTQ; (ii) can be seen as a general extended pretraining framework, meaning that the resulting model can still be utilized for any downstream task afterwards; (iii) can be applied across a wide range of quantization settings, such as different choices quantization granularity, activation quantization, and seamlessly combined with many PTQ techniques. We apply LR-QAT to LLaMA-1/2/3 and Mistral model families and validate its effectiveness on several downstream tasks. Our method outperforms common post-training quantization (PTQ) approaches and reaches the same model performance as full-model QAT at the fraction of its memory usage. Specifically, we can train a 7B LLM on a single consumer grade GPU with 24GB of memory. Our source code is available at https://github.com/qualcomm-ai-research/LR-QAT

Yelysei Bondarenko, Markus Nagel, Tijmen Blankevoort

Transformer-based architectures have become the de-facto standard models for a wide range of Natural Language Processing tasks. However, their memory footprint and high latency are prohibitive for efficient deployment and inference on resource-limited devices. In this work, we explore quantization for transformers. We show that transformers have unique quantization challenges -- namely, high dynamic activation ranges that are difficult to represent with a low bit fixed-point format. We establish that these activations contain structured outliers in the residual connections that encourage specific attention patterns, such as attending to the special separator token. To combat these challenges, we present three solutions based on post-training quantization and quantization-aware training, each with a different set of compromises for accuracy, model size, and ease of use. In particular, we introduce a novel quantization scheme -- per-embedding-group quantization. We demonstrate the effectiveness of our methods on the GLUE benchmark using BERT, establishing state-of-the-art results for post-training quantization. Finally, we show that transformer weights and embeddings can be quantized to ultra-low bit-widths, leading to significant memory savings with a minimum accuracy loss. Our source code is available at~\url{https://github.com/qualcomm-ai-research/transformer-quantization}.

Mart van Baalen, Andrey Kuzmin, Ivan Koryakovskiy et al.

In this work we show that the size versus accuracy trade-off of neural network quantization can be significantly improved by increasing the quantization dimensionality. We propose the GPTVQ method, a new fast method for post-training vector quantization (VQ) that scales well to Large Language Models (LLMs). Our method interleaves quantization of one or more columns with updates to the remaining unquantized weights, using information from the Hessian of the per-layer output reconstruction MSE. Quantization codebooks are initialized using an efficient data-aware version of the EM algorithm. The codebooks are then updated, and further compressed by using integer quantization and SVD-based compression. GPTVQ establishes a new state-of-the art in the size vs accuracy trade-offs on a wide range of LLMs such as Llama-v2 and Mistral. Furthermore, our method is efficient: on a single H100 it takes between 3 and 11 hours to process a Llamav2-70B model, depending on quantization setting. Lastly, with on-device timings for VQ decompression on a mobile CPU we show that VQ leads to improved latency compared to using a 4-bit integer format.

Tycho F. A. van der Ouderaa, Markus Nagel, Mart van Baalen et al.

State-of-the-art language models are becoming increasingly large in an effort to achieve the highest performance on large corpora of available textual data. However, the sheer size of the Transformer architectures makes it difficult to deploy models within computational, environmental or device-specific constraints. We explore data-driven compression of existing pretrained models as an alternative to training smaller models from scratch. To do so, we scale Kronecker-factored curvature approximations of the target loss landscape to large language models. In doing so, we can compute both the dynamic allocation of structures that can be removed as well as updates of remaining weights that account for the removal. We provide a general framework for unstructured, semi-structured and structured pruning and improve upon weight updates to capture more correlations between weights, while remaining computationally efficient. Experimentally, our method can prune rows and columns from a range of OPT models and Llamav2-7B by 20%-30%, with a negligible loss in performance, and achieve state-of-the-art results in unstructured and semi-structured pruning of large language models.

Andrii Skliar, Ties van Rozendaal, Romain Lepert et al.

Mixture of Experts (MoE) LLMs have recently gained attention for their ability to enhance performance by selectively engaging specialized subnetworks or "experts" for each input. However, deploying MoEs on memory-constrained devices remains challenging, particularly when generating tokens sequentially with a batch size of one, as opposed to typical high-throughput settings involving long sequences or large batches. In this work, we optimize MoE on memory-constrained devices where only a subset of expert weights fit in DRAM. We introduce a novel cache-aware routing strategy that leverages expert reuse during token generation to improve cache locality. We evaluate our approach on language modeling, MMLU, and GSM8K benchmarks and present on-device results demonstrating 2$\times$ speedups on mobile devices, offering a flexible, training-free solution to extend MoE's applicability across real-world applications.

Boris van Breugel, Yelysei Bondarenko, Paul Whatmough et al.

Large language models (LLMs) require substantial compute, and thus energy, at inference time. While quantizing weights and activations is effective at improving efficiency, naive quantization of LLMs can significantly degrade performance due to large magnitude outliers. This paper describes FPTQuant, which introduces four novel, lightweight, and expressive function-preserving transforms (FPTs) to facilitate quantization of transformers: (1) a mergeable pre-RoPE transform for queries and keys, (2) a mergeable transform for values, (3) a mergeable scaling transform within the MLP block, and (4) a cheap, dynamic scaling transform. By leveraging the equivariances and independencies inherent to canonical transformer operation, we designed these FPTs to maintain the model's function while shaping the intermediate activation distributions to be more quantization friendly. FPTQuant requires no custom kernels and adds virtually no overhead during inference. The FPTs are trained both locally to reduce outliers, and end-to-end such that the outputs of the quantized and full-precision models match. FPTQuant enables static INT4 quantization with minimal overhead and shows SOTA speed-up of up to 3.9 times over FP. Empirically, FPTQuant has an excellent accuracy-speed trade-off -- it is performing on par or exceeding most prior work and only shows slightly lower accuracy compared to a method that is up to 29% slower.

Marco Federici, Riccardo Del Chiaro, Boris van Breugel et al.

Diffusion models represent the cutting edge in image generation, but their high memory and computational demands hinder deployment on resource-constrained devices. Post-Training Quantization (PTQ) offers a promising solution by reducing the bitwidth of matrix operations. However, standard PTQ methods struggle with outliers, and achieving higher compression often requires transforming model weights and activations before quantization. In this work, we propose HadaNorm, a novel linear transformation that extends existing approaches by both normalizing channels activations and applying Hadamard transforms to effectively mitigate outliers and enable aggressive activation quantization. We demonstrate that HadaNorm consistently reduces quantization error across the various components of transformer blocks, outperforming state-of-the-art methods.

Evgenii Egorov, Hanno Ackermann, Markus Nagel et al.

Attention layers apply a sequence-to-sequence mapping whose parameters depend on the pairwise interactions of the input elements. However, without any structural assumptions, memory and compute scale quadratically with the sequence length. The two main ways to mitigate this are to introduce sparsity by ignoring a sufficient amount of pairwise interactions or to introduce recurrent dependence along them, as SSM does. Although both approaches are reasonable, they both have disadvantages. We propose a novel algorithm that combines the advantages of both concepts. Our idea is based on the efficient inversion of tree-structured matrices.

Kartikeya Bhardwaj, Nilesh Prasad Pandey, Sweta Priyadarshi et al.

Low Rank Adaptation (LoRA) has gained massive attention in the recent generative AI research. One of the main advantages of LoRA is its ability to be fused with pretrained models, adding no overhead during inference. However, from a mobile deployment standpoint, we can either avoid inference overhead in the fused mode but lose the ability to switch adapters rapidly, or suffer significant (up to 30% higher) inference latency while enabling rapid switching in the unfused mode. LoRA also exhibits concept-loss when multiple adapters are used concurrently. In this paper, we propose Sparse High Rank Adapters (SHiRA), a new paradigm which incurs no inference overhead, enables rapid switching, and significantly reduces concept-loss. Specifically, SHiRA can be trained by directly tuning only 1-2% of the base model weights while leaving others unchanged. This results in a highly sparse adapter which can be switched directly in the fused mode. We further provide theoretical and empirical insights on how high sparsity in SHiRA can aid multi-adapter fusion by reducing concept loss. Our extensive experiments on LVMs and LLMs demonstrate that finetuning only a small fraction of the parameters in the base model significantly outperforms LoRA while enabling both rapid switching and multi-adapter fusion. Finally, we provide a latency- and memory-efficient SHiRA implementation based on Parameter-Efficient Finetuning (PEFT) Library which trains at nearly the same speed as LoRA while consuming up to 16% lower peak GPU memory, thus making SHiRA easy to adopt for practical use cases. To demonstrate rapid switching benefits during inference, we show that loading SHiRA on a base model can be 5x-16x faster than LoRA fusion on a CPU.

Suraj Srinivas, Andrey Kuzmin, Markus Nagel et al.

Current methods for pruning neural network weights iteratively apply magnitude-based pruning on the model weights and re-train the resulting model to recover lost accuracy. In this work, we show that such strategies do not allow for the recovery of erroneously pruned weights. To enable weight recovery, we propose a simple strategy called \textit{cyclical pruning} which requires the pruning schedule to be periodic and allows for weights pruned erroneously in one cycle to recover in subsequent ones. Experimental results on both linear models and large-scale deep neural networks show that cyclical pruning outperforms existing pruning algorithms, especially at high sparsity ratios. Our approach is easy to tune and can be readily incorporated into existing pruning pipelines to boost performance.

Sangeetha Siddegowda, Marios Fournarakis, Markus Nagel et al.

While neural networks have advanced the frontiers in many machine learning applications, they often come at a high computational cost. Reducing the power and latency of neural network inference is vital to integrating modern networks into edge devices with strict power and compute requirements. Neural network quantization is one of the most effective ways of achieving these savings, but the additional noise it induces can lead to accuracy degradation. In this white paper, we present an overview of neural network quantization using AI Model Efficiency Toolkit (AIMET). AIMET is a library of state-of-the-art quantization and compression algorithms designed to ease the effort required for model optimization and thus drive the broader AI ecosystem towards low latency and energy-efficient inference. AIMET provides users with the ability to simulate as well as optimize PyTorch and TensorFlow models. Specifically for quantization, AIMET includes various post-training quantization (PTQ, cf. chapter 4) and quantization-aware training (QAT, cf. chapter 5) techniques that guarantee near floating-point accuracy for 8-bit fixed-point inference. We provide a practical guide to quantization via AIMET by covering PTQ and QAT workflows, code examples and practical tips that enable users to efficiently and effectively quantize models using AIMET and reap the benefits of low-bit integer inference.

Yunfan Zhang, Ties van Rozendaal, Johann Brehmer et al.

We propose a method to compress full-resolution video sequences with implicit neural representations. Each frame is represented as a neural network that maps coordinate positions to pixel values. We use a separate implicit network to modulate the coordinate inputs, which enables efficient motion compensation between frames. Together with a small residual network, this allows us to efficiently compress P-frames relative to the previous frame. We further lower the bitrate by storing the network weights with learned integer quantization. Our method, which we call implicit pixel flow (IPF), offers several simplifications over established neural video codecs: it does not require the receiver to have access to a pretrained neural network, does not use expensive interpolation-based warping operations, and does not require a separate training dataset. We demonstrate the feasibility of neural implicit compression on image and video data.

Markus Nagel, Marios Fournarakis, Rana Ali Amjad et al.

While neural networks have advanced the frontiers in many applications, they often come at a high computational cost. Reducing the power and latency of neural network inference is key if we want to integrate modern networks into edge devices with strict power and compute requirements. Neural network quantization is one of the most effective ways of achieving these savings but the additional noise it induces can lead to accuracy degradation. In this white paper, we introduce state-of-the-art algorithms for mitigating the impact of quantization noise on the network's performance while maintaining low-bit weights and activations. We start with a hardware motivated introduction to quantization and then consider two main classes of algorithms: Post-Training Quantization (PTQ) and Quantization-Aware-Training (QAT). PTQ requires no re-training or labelled data and is thus a lightweight push-button approach to quantization. In most cases, PTQ is sufficient for achieving 8-bit quantization with close to floating-point accuracy. QAT requires fine-tuning and access to labeled training data but enables lower bit quantization with competitive results. For both solutions, we provide tested pipelines based on existing literature and extensive experimentation that lead to state-of-the-art performance for common deep learning models and tasks.

Marios Fournarakis, Markus Nagel

Quantization techniques applied to the inference of deep neural networks have enabled fast and efficient execution on resource-constraint devices. The success of quantization during inference has motivated the academic community to explore fully quantized training, i.e. quantizing back-propagation as well. However, effective gradient quantization is still an open problem. Gradients are unbounded and their distribution changes significantly during training, which leads to the need for dynamic quantization. As we show, dynamic quantization can lead to significant memory overhead and additional data traffic slowing down training. We propose a simple alternative to dynamic quantization, in-hindsight range estimation, that uses the quantization ranges estimated on previous iterations to quantize the present. Our approach enables fast static quantization of gradients and activations while requiring only minimal hardware support from the neural network accelerator to keep track of output statistics in an online fashion. It is intended as a drop-in replacement for estimating quantization ranges and can be used in conjunction with other advances in quantized training. We compare our method to existing methods for range estimation from the quantized training literature and demonstrate its effectiveness with a range of architectures, including MobileNetV2, on image classification benchmarks (Tiny ImageNet & ImageNet).

Mart van Baalen, Christos Louizos, Markus Nagel et al.

We introduce Bayesian Bits, a practical method for joint mixed precision quantization and pruning through gradient based optimization. Bayesian Bits employs a novel decomposition of the quantization operation, which sequentially considers doubling the bit width. At each new bit width, the residual error between the full precision value and the previously rounded value is quantized. We then decide whether or not to add this quantized residual error for a higher effective bit width and lower quantization noise. By starting with a power-of-two bit width, this decomposition will always produce hardware-friendly configurations, and through an additional 0-bit option, serves as a unified view of pruning and quantization. Bayesian Bits then introduces learnable stochastic gates, which collectively control the bit width of the given tensor. As a result, we can obtain low bit solutions by performing approximate inference over the gates, with prior distributions that encourage most of them to be switched off. We experimentally validate our proposed method on several benchmark datasets and show that we can learn pruned, mixed precision networks that provide a better trade-off between accuracy and efficiency than their static bit width equivalents.

Markus Nagel, Rana Ali Amjad, Mart van Baalen et al.

When quantizing neural networks, assigning each floating-point weight to its nearest fixed-point value is the predominant approach. We find that, perhaps surprisingly, this is not the best we can do. In this paper, we propose AdaRound, a better weight-rounding mechanism for post-training quantization that adapts to the data and the task loss. AdaRound is fast, does not require fine-tuning of the network, and only uses a small amount of unlabelled data. We start by theoretically analyzing the rounding problem for a pre-trained neural network. By approximating the task loss with a Taylor series expansion, the rounding task is posed as a quadratic unconstrained binary optimization problem. We simplify this to a layer-wise local loss and propose to optimize this loss with a soft relaxation. AdaRound not only outperforms rounding-to-nearest by a significant margin but also establishes a new state-of-the-art for post-training quantization on several networks and tasks. Without fine-tuning, we can quantize the weights of Resnet18 and Resnet50 to 4 bits while staying within an accuracy loss of 1%.

Yash Bhalgat, Jinwon Lee, Markus Nagel et al.

Unlike ReLU, newer activation functions (like Swish, H-swish, Mish) that are frequently employed in popular efficient architectures can also result in negative activation values, with skewed positive and negative ranges. Typical learnable quantization schemes [PACT, LSQ] assume unsigned quantization for activations and quantize all negative activations to zero which leads to significant loss in performance. Naively using signed quantization to accommodate these negative values requires an extra sign bit which is expensive for low-bit (2-, 3-, 4-bit) quantization. To solve this problem, we propose LSQ+, a natural extension of LSQ, wherein we introduce a general asymmetric quantization scheme with trainable scale and offset parameters that can learn to accommodate the negative activations. Gradient-based learnable quantization schemes also commonly suffer from high instability or variance in the final training performance, hence requiring a great deal of hyper-parameter tuning to reach a satisfactory performance. LSQ+ alleviates this problem by using an MSE-based initialization scheme for the quantization parameters. We show that this initialization leads to significantly lower variance in final performance across multiple training runs. Overall, LSQ+ shows state-of-the-art results for EfficientNet and MixNet and also significantly outperforms LSQ for low-bit quantization of neural nets with Swish activations (e.g.: 1.8% gain with W4A4 quantization and upto 5.6% gain with W2A2 quantization of EfficientNet-B0 on ImageNet dataset). To the best of our knowledge, ours is the first work to quantize such architectures to extremely low bit-widths.

Andrey Kuzmin, Markus Nagel, Saurabh Pitre et al.

The success of deep neural networks in many real-world applications is leading to new challenges in building more efficient architectures. One effective way of making networks more efficient is neural network compression. We provide an overview of existing neural network compression methods that can be used to make neural networks more efficient by changing the architecture of the network. First, we introduce a new way to categorize all published compression methods, based on the amount of data and compute needed to make the methods work in practice. These are three 'levels of compression solutions'. Second, we provide a taxonomy of tensor factorization based and probabilistic compression methods. Finally, we perform an extensive evaluation of different compression techniques from the literature for models trained on ImageNet. We show that SVD and probabilistic compression or pruning methods are complementary and give the best results of all the considered methods. We also provide practical ways to combine them.

Markus Nagel, Mart van Baalen, Tijmen Blankevoort et al.

We introduce a data-free quantization method for deep neural networks that does not require fine-tuning or hyperparameter selection. It achieves near-original model performance on common computer vision architectures and tasks. 8-bit fixed-point quantization is essential for efficient inference on modern deep learning hardware. However, quantizing models to run in 8-bit is a non-trivial task, frequently leading to either significant performance reduction or engineering time spent on training a network to be amenable to quantization. Our approach relies on equalizing the weight ranges in the network by making use of a scale-equivariance property of activation functions. In addition the method corrects biases in the error that are introduced during quantization. This improves quantization accuracy performance, and can be applied to many common computer vision architectures with a straight forward API call. For common architectures, such as the MobileNet family, we achieve state-of-the-art quantized model performance. We further show that the method also extends to other computer vision architectures and tasks such as semantic segmentation and object detection.