Zhiwei Hao

Zhiwei Hao, Jianyuan Guo, Kai Han et al.

Knowledge distillation~(KD) has proven to be a highly effective approach for enhancing model performance through a teacher-student training scheme. However, most existing distillation methods are designed under the assumption that the teacher and student models belong to the same model family, particularly the hint-based approaches. By using centered kernel alignment (CKA) to compare the learned features between heterogeneous teacher and student models, we observe significant feature divergence. This divergence illustrates the ineffectiveness of previous hint-based methods in cross-architecture distillation. To tackle the challenge in distilling heterogeneous models, we propose a simple yet effective one-for-all KD framework called OFA-KD, which significantly improves the distillation performance between heterogeneous architectures. Specifically, we project intermediate features into an aligned latent space such as the logits space, where architecture-specific information is discarded. Additionally, we introduce an adaptive target enhancement scheme to prevent the student from being disturbed by irrelevant information. Extensive experiments with various architectures, including CNN, Transformer, and MLP, demonstrate the superiority of our OFA-KD framework in enabling distillation between heterogeneous architectures. Specifically, when equipped with our OFA-KD, the student models achieve notable performance improvements, with a maximum gain of 8.0% on the CIFAR-100 dataset and 0.7% on the ImageNet-1K dataset. PyTorch code and checkpoints can be found at https://github.com/Hao840/OFAKD.

Zhiwei Hao, Guanyu Xu, Yong Luo et al.

Recently, deploying deep neural network (DNN) models via collaborative inference, which splits a pre-trained model into two parts and executes them on user equipment (UE) and edge server respectively, becomes attractive. However, the large intermediate feature of DNN impedes flexible decoupling, and existing approaches either focus on the single UE scenario or simply define tasks considering the required CPU cycles, but ignore the indivisibility of a single DNN layer. In this paper, we study the multi-agent collaborative inference scenario, where a single edge server coordinates the inference of multiple UEs. Our goal is to achieve fast and energy-efficient inference for all UEs. To achieve this goal, we first design a lightweight autoencoder-based method to compress the large intermediate feature. Then we define tasks according to the inference overhead of DNNs and formulate the problem as a Markov decision process (MDP). Finally, we propose a multi-agent hybrid proximal policy optimization (MAHPPO) algorithm to solve the optimization problem with a hybrid action space. We conduct extensive experiments with different types of networks, and the results show that our method can reduce up to 56\% of inference latency and save up to 72\% of energy consumption.

Zhiwei Hao, Yong Luo, Zhi Wang et al.

Recently, the compression and deployment of powerful deep neural networks (DNNs) on resource-limited edge devices to provide intelligent services have become attractive tasks. Although knowledge distillation (KD) is a feasible solution for compression, its requirement on the original dataset raises privacy concerns. In addition, it is common to integrate multiple pretrained models to achieve satisfactory performance. How to compress multiple models into a tiny model is challenging, especially when the original data are unavailable. To tackle this challenge, we propose a framework termed collaborative data-free knowledge distillation via multi-level feature sharing (CDFKD-MFS), which consists of a multi-header student module, an asymmetric adversarial data-free KD module, and an attention-based aggregation module. In this framework, the student model equipped with a multi-level feature-sharing structure learns from multiple teacher models and is trained together with a generator in an asymmetric adversarial manner. When some real samples are available, the attention module adaptively aggregates predictions of the student headers, which can further improve performance. We conduct extensive experiments on three popular computer visual datasets. In particular, compared with the most competitive alternative, the accuracy of the proposed framework is 1.18\% higher on the CIFAR-100 dataset, 1.67\% higher on the Caltech-101 dataset, and 2.99\% higher on the mini-ImageNet dataset.

Guanyu Xu, Zhiwei Hao, Yong Luo et al.

Recent years have witnessed the great success of vision transformer (ViT), which has achieved state-of-the-art performance on multiple computer vision benchmarks. However, ViT models suffer from vast amounts of parameters and high computation cost, leading to difficult deployment on resource-constrained edge devices. Existing solutions mostly compress ViT models to a compact model but still cannot achieve real-time inference. To tackle this issue, we propose to explore the divisibility of transformer structure, and decompose the large ViT into multiple small models for collaborative inference at edge devices. Our objective is to achieve fast and energy-efficient collaborative inference while maintaining comparable accuracy compared with large ViTs. To this end, we first propose a collaborative inference framework termed DeViT to facilitate edge deployment by decomposing large ViTs. Subsequently, we design a decomposition-and-ensemble algorithm based on knowledge distillation, termed DEKD, to fuse multiple small decomposed models while dramatically reducing communication overheads, and handle heterogeneous models by developing a feature matching module to promote the imitations of decomposed models from the large ViT. Extensive experiments for three representative ViT backbones on four widely-used datasets demonstrate our method achieves efficient collaborative inference for ViTs and outperforms existing lightweight ViTs, striking a good trade-off between efficiency and accuracy. For example, our DeViTs improves end-to-end latency by 2.89$\times$ with only 1.65% accuracy sacrifice using CIFAR-100 compared to the large ViT, ViT-L/16, on the GPU server. DeDeiTs surpasses the recent efficient ViT, MobileViT-S, by 3.54% in accuracy on ImageNet-1K, while running 1.72$\times$ faster and requiring 55.28% lower energy consumption on the edge device.

Zhiwei Hao, Wenrong Jiang, Nan Li et al.

Given a polynomial system f associated with a simple multiple zero x of multiplicity μ, we give a computable lower bound on the minimal distance between the simple multiple zero x and other zeros of f. If x is only given with limited accuracy, we propose a numerical criterion that f is certified to have μ zeros (counting multiplicities) in a small ball around x. Furthermore, for simple double zeros and simple triple zeros whose Jacobian is of normalized form, we define modified Newton iterations and prove the quantified quadratic convergence when the starting point is close to the exact simple multiple zero. For simple multiple zeros of arbitrary multiplicity whose Jacobian matrix may not have a normalized form, we perform unitary transformations and modified Newton iterations, and prove its non-quantified quadratic convergence and its quantified convergence for simple triple zeros.

Zhenhua Liu, Zhiwei Hao, Kai Han et al.

Compact neural networks are specially designed for applications on edge devices with faster inference speed yet modest performance. However, training strategies of compact models are borrowed from that of conventional models at present, which ignores their difference in model capacity and thus may impede the performance of compact models. In this paper, by systematically investigating the impact of different training ingredients, we introduce a strong training strategy for compact models. We find that the appropriate designs of re-parameterization and knowledge distillation are crucial for training high-performance compact models, while some commonly used data augmentations for training conventional models, such as Mixup and CutMix, lead to worse performance. Our experiments on ImageNet-1K dataset demonstrate that our specialized training strategy for compact models is applicable to various architectures, including GhostNetV2, MobileNetV2 and ShuffleNetV2. Specifically, equipped with our strategy, GhostNetV3 1.3$\times$ achieves a top-1 accuracy of 79.1% with only 269M FLOPs and a latency of 14.46ms on mobile devices, surpassing its ordinarily trained counterpart by a large margin. Moreover, our observation can also be extended to object detection scenarios. PyTorch code and checkpoints can be found at https://github.com/huawei-noah/Efficient-AI-Backbones/tree/master/ghostnetv3_pytorch.

Chengcheng Wang, Zhiwei Hao, Yehui Tang et al.

Diffusion-based super-resolution (SR) models have recently garnered significant attention due to their potent restoration capabilities. But conventional diffusion models perform noise sampling from a single distribution, constraining their ability to handle real-world scenes and complex textures across semantic regions. With the success of segment anything model (SAM), generating sufficiently fine-grained region masks can enhance the detail recovery of diffusion-based SR model. However, directly integrating SAM into SR models will result in much higher computational cost. In this paper, we propose the SAM-DiffSR model, which can utilize the fine-grained structure information from SAM in the process of sampling noise to improve the image quality without additional computational cost during inference. In the process of training, we encode structural position information into the segmentation mask from SAM. Then the encoded mask is integrated into the forward diffusion process by modulating it to the sampled noise. This adjustment allows us to independently adapt the noise mean within each corresponding segmentation area. The diffusion model is trained to estimate this modulated noise. Crucially, our proposed framework does NOT change the reverse diffusion process and does NOT require SAM at inference. Experimental results demonstrate the effectiveness of our proposed method, showcasing superior performance in suppressing artifacts, and surpassing existing diffusion-based methods by 0.74 dB at the maximum in terms of PSNR on DIV2K dataset. The code and dataset are available at https://github.com/lose4578/SAM-DiffSR.

Jianyuan Guo, Zhiwei Hao, Chengcheng Wang et al.

Training general-purpose vision models on purely sequential visual data, eschewing linguistic inputs, has heralded a new frontier in visual understanding. These models are intended to not only comprehend but also seamlessly transit to out-of-domain tasks. However, current endeavors are hamstrung by an over-reliance on colossal models, exemplified by models with upwards of 3B parameters, and the necessity for an extensive corpus of visual data, often comprising a staggering 400B tokens. In this paper, we delve into the development of an efficient, autoregression-based vision model, innovatively architected to operate on a limited dataset. We meticulously demonstrate how this model achieves proficiency in a spectrum of visual tasks spanning both high-level and low-level semantic understanding during the testing phase. Our empirical evaluations underscore the model's agility in adapting to various tasks, heralding a significant reduction in the parameter footprint, and a marked decrease in training data requirements, thereby paving the way for more sustainable and accessible advancements in the field of generalist vision models. The code is available at https://github.com/ggjy/DeLVM.

Zhiwei Hao, Jianyuan Guo, Li Shen et al.

Large language models (LLMs) have achieved impressive performance across various domains. However, the substantial hardware resources required for their training present a significant barrier to efficiency and scalability. To mitigate this challenge, low-precision training techniques have been widely adopted, leading to notable advancements in training efficiency. Despite these gains, low-precision training involves several components$\unicode{x2013}$such as weights, activations, and gradients$\unicode{x2013}$each of which can be represented in different numerical formats. The resulting diversity has created a fragmented landscape in low-precision training research, making it difficult for researchers to gain a unified overview of the field. This survey provides a comprehensive review of existing low-precision training methods. To systematically organize these approaches, we categorize them into three primary groups based on their underlying numerical formats, which is a key factor influencing hardware compatibility, computational efficiency, and ease of reference for readers. The categories are: (1) fixed-point and integer-based methods, (2) floating-point-based methods, and (3) customized format-based methods. Additionally, we discuss quantization-aware training approaches, which share key similarities with low-precision training during forward propagation. Finally, we highlight several promising research directions to advance this field. A collection of papers discussed in this survey is provided in https://github.com/Hao840/Awesome-Low-Precision-Training.

Zhiwei Hao, Jianyuan Guo, Li Shen et al.

Recent advancements in multimodal fusion have witnessed the remarkable success of vision-language (VL) models, which excel in various multimodal applications such as image captioning and visual question answering. However, building VL models requires substantial hardware resources, where efficiency is restricted by two key factors: the extended input sequence of the language model with vision features demands more computational operations, and a large number of additional learnable parameters increase memory complexity. These challenges significantly restrict the broader applicability of such models. To bridge this gap, we propose ADEM-VL, an efficient vision-language method that tunes VL models based on pretrained large language models (LLMs) by adopting a parameter-free cross-attention mechanism for similarity measurements in multimodal fusion. This approach only requires embedding vision features into the language space, significantly reducing the number of trainable parameters and accelerating both training and inference speeds. To enhance representation learning in fusion module, we introduce an efficient multiscale feature generation scheme that requires only a single forward pass through the vision encoder. Moreover, we propose an adaptive fusion scheme that dynamically discards less relevant visual information for each text token based on its attention score. This ensures that the fusion process prioritizes the most pertinent visual features. With experiments on various tasks including visual question answering, image captioning, and instruction-following, we demonstrate that our framework outperforms existing approaches. Specifically, our method surpasses existing methods by an average accuracy of 0.77% on ScienceQA dataset, with reduced training and inference latency, demonstrating the superiority of our framework. The code is available at https://github.com/Hao840/ADEM-VL.

Zhongyu Xiao, Zhiwei Hao, Jianyuan Guo et al.

Diffusion Large Language Models (dLLMs) offer a compelling paradigm for natural language generation, leveraging parallel decoding and bidirectional attention to achieve superior global coherence compared to autoregressive models. While recent works have accelerated inference via KV cache reuse or heuristic decoding, they overlook the intrinsic inefficiencies within the block-wise diffusion process. Specifically, they suffer from spatial redundancy by modeling informative-sparse suffix regions uniformly and temporal inefficiency by applying fixed denoising schedules across all the decoding process. To address this, we propose Streaming-dLLM, a training-free framework that streamlines inference across both spatial and temporal dimensions. Spatially, we introduce attenuation guided suffix modeling to approximate the full context by pruning redundant mask tokens. Temporally, we employ a dynamic confidence aware strategy with an early exit mechanism, allowing the model to skip unnecessary iterations for converged tokens. Extensive experiments show that Streaming-dLLM achieves up to 68.2X speedup while maintaining generation quality, highlighting its effectiveness in diffusion decoding. The code is available at https://github.com/xiaoshideta/Streaming-dLLM.

Zhiwei Hao, Jianyuan Guo, Kai Han et al.

The tremendous success of large models trained on extensive datasets demonstrates that scale is a key ingredient in achieving superior results. Therefore, the reflection on the rationality of designing knowledge distillation (KD) approaches for limited-capacity architectures solely based on small-scale datasets is now deemed imperative. In this paper, we identify the \emph{small data pitfall} that presents in previous KD methods, which results in the underestimation of the power of vanilla KD framework on large-scale datasets such as ImageNet-1K. Specifically, we show that employing stronger data augmentation techniques and using larger datasets can directly decrease the gap between vanilla KD and other meticulously designed KD variants. This highlights the necessity of designing and evaluating KD approaches in the context of practical scenarios, casting off the limitations of small-scale datasets. Our investigation of the vanilla KD and its variants in more complex schemes, including stronger training strategies and different model capacities, demonstrates that vanilla KD is elegantly simple but astonishingly effective in large-scale scenarios. Without bells and whistles, we obtain state-of-the-art ResNet-50, ViT-S, and ConvNeXtV2-T models for ImageNet, which achieve 83.1\%, 84.3\%, and 85.0\% top-1 accuracy, respectively. PyTorch code and checkpoints can be found at https://github.com/Hao840/vanillaKD.

Jiawei Hao, Zhiwei Hao, Jianyuan Guo et al.

Mixture-of-Experts (MoE) based Large Language Models (LLMs) have demonstrated impressive performance and computational efficiency. However, their deployment is often constrained by substantial memory demands, primarily due to the need to load numerous expert modules. While existing expert compression techniques like pruning or merging attempt to mitigate this, they often suffer from irreversible knowledge loss or high training overhead. In this paper, we propose a novel expert compression paradigm termed expert replacing, which replaces redundant experts with parameter-efficient modules and recovers their capabilities with low training costs. We find that even a straightforward baseline of this paradigm yields promising performance. Building on this foundation, we introduce LightMoE, a framework that enhances the paradigm by introducing adaptive expert selection, hierarchical expert construction, and an annealed recovery strategy. Experimental results show that LightMoE matches the performance of LoRA fine-tuning at a 30% compression ratio. Even under a more aggressive 50% compression rate, it outperforms existing methods and achieves average performance improvements of 5.6% across five diverse tasks. These findings demonstrate that LightMoE strikes a superior balance among memory efficiency, training efficiency, and model performance.

Zhiwei Hao, Jianyuan Guo, Li Shen et al.

Recent advancements in vision transformers (ViTs) have demonstrated that larger models often achieve superior performance. However, training these models remains computationally intensive and costly. To address this challenge, we introduce ScaleNet, an efficient approach for scaling ViT models. Unlike conventional training from scratch, ScaleNet facilitates rapid model expansion with negligible increases in parameters, building on existing pretrained models. This offers a cost-effective solution for scaling up ViTs. Specifically, ScaleNet achieves model expansion by inserting additional layers into pretrained ViTs, utilizing layer-wise weight sharing to maintain parameters efficiency. Each added layer shares its parameter tensor with a corresponding layer from the pretrained model. To mitigate potential performance degradation due to shared weights, ScaleNet introduces a small set of adjustment parameters for each layer. These adjustment parameters are implemented through parallel adapter modules, ensuring that each instance of the shared parameter tensor remains distinct and optimized for its specific function. Experiments on the ImageNet-1K dataset demonstrate that ScaleNet enables efficient expansion of ViT models. With a 2$\times$ depth-scaled DeiT-Base model, ScaleNet achieves a 7.42% accuracy improvement over training from scratch while requiring only one-third of the training epochs, highlighting its efficiency in scaling ViTs. Beyond image classification, our method shows significant potential for application in downstream vision areas, as evidenced by the validation in object detection task.

Guanyu Xu, Zhiwei Hao, Li Shen et al.

The impressive performance of transformer models has sparked the deployment of intelligent applications on resource-constrained edge devices. However, ensuring high-quality service for real-time edge systems is a significant challenge due to the considerable computational demands and resource requirements of these models. Existing strategies typically either offload transformer computations to other devices or directly deploy compressed models on individual edge devices. These strategies, however, result in either considerable communication overhead or suboptimal trade-offs between accuracy and efficiency. To tackle these challenges, we propose a collaborative inference system for general transformer models, termed CoFormer. The central idea behind CoFormer is to exploit the divisibility and integrability of transformer. An off-the-shelf large transformer can be decomposed into multiple smaller models for distributed inference, and their intermediate results are aggregated to generate the final output. We formulate an optimization problem to minimize both inference latency and accuracy degradation under heterogeneous hardware constraints. DeBo algorithm is proposed to first solve the optimization problem to derive the decomposition policy, and then progressively calibrate decomposed models to restore performance. We demonstrate the capability to support a wide range of transformer models on heterogeneous edge devices, achieving up to 3.1$\times$ inference speedup with large transformer models. Notably, CoFormer enables the efficient inference of GPT2-XL with 1.6 billion parameters on edge devices, reducing memory requirements by 76.3\%. CoFormer can also reduce energy consumption by approximately 40\% while maintaining satisfactory inference performance.

Juntong Fan, Zhiwei Hao, Jianqiang Shen et al.

Due to the excellent performance in yielding high-quality, zero-shot segmentation, Segment Anything Model (SAM) and its variants have been widely applied in diverse scenarios such as healthcare and intelligent manufacturing. Therefore, effectively compressing SAMs has become an increasingly pressing practical need. In this study, we propose Birkhoff, a novel data-free compression algorithm for SAM and its variants. Unlike quantization, pruning, distillation, and other compression methods, Birkhoff embodies versatility across model types, agility in deployment, faithfulness to the original model, and compactness in model size. Specifically, Birkhoff introduces a novel compression algorithm: Hyper-Compression, whose core principle is to find a dense trajectory to turn a high-dimensional parameter vector into a low-dimensional scalar. Furthermore, Birkhoff designs a dedicated linear layer operator, HyperLinear, to fuse decompression and matrix multiplication to significantly accelerate inference of the compressed SAMs. Extensive experiments on 18 SAMs in the COCO, LVIS, and SA-1B datasets show that Birkhoff performs consistently and competitively in compression time, compression ratio, post-compression performance, and inference speed. For example, Birkhoff can achieve a compression ratio of 5.17x on SAM2-B, with less than 1% performance drop without using any fine-tuning data. Moreover, the compression is finished within 60 seconds for all models.

Zhiwei Hao, Jianyuan Guo, Ding Jia et al.

In the past few years, transformers have achieved promising performances on various computer vision tasks. Unfortunately, the immense inference overhead of most existing vision transformers withholds their from being deployed on edge devices such as cell phones and smart watches. Knowledge distillation is a widely used paradigm for compressing cumbersome architectures via transferring information to a compact student. However, most of them are designed for convolutional neural networks (CNNs), which do not fully investigate the character of vision transformer (ViT). In this paper, we utilize the patch-level information and propose a fine-grained manifold distillation method. Specifically, we train a tiny student model to match a pre-trained teacher model in the patch-level manifold space. Then, we decouple the manifold matching loss into three terms with careful design to further reduce the computational costs for the patch relationship. Equipped with the proposed method, a DeiT-Tiny model containing 5M parameters achieves 76.5% top-1 accuracy on ImageNet-1k, which is +2.0% higher than previous distillation approaches. Transfer learning results on other classification benchmarks and downstream vision tasks also demonstrate the superiority of our method over the state-of-the-art algorithms.