Searcharxiv⌕ Search

arXiv subjects

Song Han

Publications and source records attributed to Song Han.

At least 163 records · Page 9Linked to original sources
arXiv2023

Lightening-Transformer: A Dynamically-operated Optically-interconnected Photonic Transformer Accelerator

The wide adoption and significant computing resource of attention-based transformers, e.g., Vision Transformers and large language models (LLM), have driven the demand for efficient hardware accelerators. There is a growing interest in exploring photonics as an alternative technology to digital electronics due to its high energy efficiency and ultra-fast processing speed. Photonic accelerators have shown promising results for CNNs, which mainly rely on weight-static linear operations. However, they encounter issues when efficiently supporting Transformer architectures, questioning the applicability of photonics to advanced ML tasks. The primary hurdle lies in their inefficiency in handling unique workloads in Transformers, i.e., dynamic and full-range tensor multiplication. In this work, we propose Lightening-Transformer, the first light-empowered, high-performance, and energy-efficient photonic Transformer accelerator. To overcome prior designs' fundamental limitations, we introduce a novel dynamically-operated photonic tensor core, DPTC, a crossbar array of interference-based optical vector dot-product engines supporting highly parallel, dynamic, and full-range matrix multiplication. Furthermore, we design a dedicated accelerator that integrates our novel photonic computing cores with photonic interconnects for inter-core data broadcast, fully unleashing the power of optics. Comprehensive evaluations show that ours achieves >2.6x energy and >12x latency reductions compared to prior photonic accelerators and delivers the lowest energy cost and 2 to 3 orders of magnitude lower energy-delay product compared to electronic Transformer accelerators, all while maintaining digital-comparable accuracy. Our work highlights the immense potential of photonics for advanced ML workloads, such as Transformer-backboned LLM. Our work is available at https://github.com/zhuhanqing/Lightening-Transformer.

cs.ET↗
arXiv2023

RobustState: Boosting Fidelity of Quantum State Preparation via Noise-Aware Variational Training

Quantum state preparation, a crucial subroutine in quantum computing, involves generating a target quantum state from initialized qubits. Arbitrary state preparation algorithms can be broadly categorized into arithmetic decomposition (AD) and variational quantum state preparation (VQSP). AD employs a predefined procedure to decompose the target state into a series of gates, whereas VQSP iteratively tunes ansatz parameters to approximate target state. VQSP is particularly apt for Noisy-Intermediate Scale Quantum (NISQ) machines due to its shorter circuits. However, achieving noise-robust parameter optimization still remains challenging. We present RobustState, a novel VQSP training methodology that combines high robustness with high training efficiency. The core idea involves utilizing measurement outcomes from real machines to perform back-propagation through classical simulators, thus incorporating real quantum noise into gradient calculations. RobustState serves as a versatile, plug-and-play technique applicable for training parameters from scratch or fine-tuning existing parameters to enhance fidelity on target machines. It is adaptable to various ansatzes at both gate and pulse levels and can even benefit other variational algorithms, such as variational unitary synthesis. Comprehensive evaluation of RobustState on state preparation tasks for 4 distinct quantum algorithms using 10 real quantum machines demonstrates a coherent error reduction of up to 7.1 $\times$ and state fidelity improvement of up to 96\% and 81\% for 4-Q and 5-Q states, respectively. On average, RobustState improves fidelity by 50\% and 72\% for 4-Q and 5-Q states compared to baseline approaches.

quant-ph↗
arXiv2023

Transformer-QEC: Quantum Error Correction Code Decoding with Transferable Transformers

Quantum computing has the potential to solve problems that are intractable for classical systems, yet the high error rates in contemporary quantum devices often exceed tolerable limits for useful algorithm execution. Quantum Error Correction (QEC) mitigates this by employing redundancy, distributing quantum information across multiple data qubits and utilizing syndrome qubits to monitor their states for errors. The syndromes are subsequently interpreted by a decoding algorithm to identify and correct errors in the data qubits. This task is complex due to the multiplicity of error sources affecting both data and syndrome qubits as well as syndrome extraction operations. Additionally, identical syndromes can emanate from different error sources, necessitating a decoding algorithm that evaluates syndromes collectively. Although machine learning (ML) decoders such as multi-layer perceptrons (MLPs) and convolutional neural networks (CNNs) have been proposed, they often focus on local syndrome regions and require retraining when adjusting for different code distances. We introduce a transformer-based QEC decoder which employs self-attention to achieve a global receptive field across all input syndromes. It incorporates a mixed loss training approach, combining both local physical error and global parity label losses. Moreover, the transformer architecture's inherent adaptability to variable-length inputs allows for efficient transfer learning, enabling the decoder to adapt to varying code distances without retraining. Evaluation on six code distances and ten different error configurations demonstrates that our model consistently outperforms non-ML decoders, such as Union Find (UF) and Minimum Weight Perfect Matching (MWPM), and other ML decoders, thereby achieving best logical error rates. Moreover, the transfer learning can save over 10x of training cost.

quant-ph↗
arXiv2023

Machine learning's own Industrial Revolution

Machine learning is expected to enable the next Industrial Revolution. However, lacking standardized and automated assembly networks, ML faces significant challenges to meet ever-growing enterprise demands and empower broad industries. In the Perspective, we argue that ML needs to first complete its own Industrial Revolution, elaborate on how to best achieve its goals, and discuss new opportunities to enable rapid translation from ML's innovation frontier to mass production and utilization.

cs.LG↗
arXiv2023

PockEngine: Sparse and Efficient Fine-tuning in a Pocket

On-device learning and efficient fine-tuning enable continuous and privacy-preserving customization (e.g., locally fine-tuning large language models on personalized data). However, existing training frameworks are designed for cloud servers with powerful accelerators (e.g., GPUs, TPUs) and lack the optimizations for learning on the edge, which faces challenges of resource limitations and edge hardware diversity. We introduce PockEngine: a tiny, sparse and efficient engine to enable fine-tuning on various edge devices. PockEngine supports sparse backpropagation: it prunes the backward graph and sparsely updates the model with measured memory saving and latency reduction while maintaining the model quality. Secondly, PockEngine is compilation first: the entire training graph (including forward, backward and optimization steps) is derived at compile-time, which reduces the runtime overhead and brings opportunities for graph transformations. PockEngine also integrates a rich set of training graph optimizations, thus can further accelerate the training cost, including operator reordering and backend switching. PockEngine supports diverse applications, frontends and hardware backends: it flexibly compiles and tunes models defined in PyTorch/TensorFlow/Jax and deploys binaries to mobile CPU/GPU/DSPs. We evaluated PockEngine on both vision models and large language models. PockEngine achieves up to 15 $\times$ speedup over off-the-shelf TensorFlow (Raspberry Pi), 5.6 $\times$ memory saving back-propagation (Jetson AGX Orin). Remarkably, PockEngine enables fine-tuning LLaMav2-7B on NVIDIA Jetson AGX Orin at 550 tokens/s, 7.9$\times$ faster than the PyTorch.

cs.LG↗
arXiv2023

TorchSparse++: Efficient Training and Inference Framework for Sparse Convolution on GPUs

Sparse convolution plays a pivotal role in emerging workloads, including point cloud processing in AR/VR, autonomous driving, and graph understanding in recommendation systems. Since the computation pattern is sparse and irregular, specialized high-performance kernels are required. Existing GPU libraries offer two dataflow types for sparse convolution. The gather-GEMM-scatter dataflow is easy to implement but not optimal in performance, while the dataflows with overlapped computation and memory access (e.g.implicit GEMM) are highly performant but have very high engineering costs. In this paper, we introduce TorchSparse++, a new GPU library that achieves the best of both worlds. We create a highly efficient Sparse Kernel Generator that generates performant sparse convolution kernels at less than one-tenth of the engineering cost of the current state-of-the-art system. On top of this, we design the Sparse Autotuner, which extends the design space of existing sparse convolution libraries and searches for the best dataflow configurations for training and inference workloads. Consequently, TorchSparse++ achieves 2.9x, 3.3x, 2.2x and 1.7x measured end-to-end speedup on an NVIDIA A100 GPU over state-of-the-art MinkowskiEngine, SpConv 1.2, TorchSparse and SpConv v2 in inference; and is 1.2-1.3x faster than SpConv v2 in mixed precision training across seven representative autonomous driving benchmarks. It also seamlessly supports graph convolutions, achieving 2.6-7.6x faster inference speed compared with state-of-the-art graph deep learning libraries.

cs.DC↗
arXiv2023

Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models

During image editing, existing deep generative models tend to re-synthesize the entire output from scratch, including the unedited regions. This leads to a significant waste of computation, especially for minor editing operations. In this work, we present Spatially Sparse Inference (SSI), a general-purpose technique that selectively performs computation for edited regions and accelerates various generative models, including both conditional GANs and diffusion models. Our key observation is that users prone to gradually edit the input image. This motivates us to cache and reuse the feature maps of the original image. Given an edited image, we sparsely apply the convolutional filters to the edited regions while reusing the cached features for the unedited areas. Based on our algorithm, we further propose Sparse Incremental Generative Engine (SIGE) to convert the computation reduction to latency reduction on off-the-shelf hardware. With about $1\%$-area edits, SIGE accelerates DDPM by $3.0\times$ on NVIDIA RTX 3090 and $4.6\times$ on Apple M1 Pro GPU, Stable Diffusion by $7.2\times$ on 3090, and GauGAN by $5.6\times$ on 3090 and $5.2\times$ on M1 Pro GPU. Compared to our conference version, we extend SIGE to accommodate attention layers and apply it to Stable Diffusion. Additionally, we offer support for Apple M1 Pro GPU and include more results with large and sequential edits.

cs.CV↗
arXiv2023

FlatFormer: Flattened Window Attention for Efficient Point Cloud Transformer

Transformer, as an alternative to CNN, has been proven effective in many modalities (e.g., texts and images). For 3D point cloud transformers, existing efforts focus primarily on pushing their accuracy to the state-of-the-art level. However, their latency lags behind sparse convolution-based models (3x slower), hindering their usage in resource-constrained, latency-sensitive applications (such as autonomous driving). This inefficiency comes from point clouds' sparse and irregular nature, whereas transformers are designed for dense, regular workloads. This paper presents FlatFormer to close this latency gap by trading spatial proximity for better computational regularity. We first flatten the point cloud with window-based sorting and partition points into groups of equal sizes rather than windows of equal shapes. This effectively avoids expensive structuring and padding overheads. We then apply self-attention within groups to extract local features, alternate sorting axis to gather features from different directions, and shift windows to exchange features across groups. FlatFormer delivers state-of-the-art accuracy on Waymo Open Dataset with 4.6x speedup over (transformer-based) SST and 1.4x speedup over (sparse convolutional) CenterPoint. This is the first point cloud transformer that achieves real-time performance on edge GPUs and is faster than sparse convolutional methods while achieving on-par or even superior accuracy on large-scale benchmarks.

cs.CV↗
arXiv2023

Retrospective: EIE: Efficient Inference Engine on Sparse and Compressed Neural Network

EIE proposed to accelerate pruned and compressed neural networks, exploiting weight sparsity, activation sparsity, and 4-bit weight-sharing in neural network accelerators. Since published in ISCA'16, it opened a new design space to accelerate pruned and sparse neural networks and spawned many algorithm-hardware co-designs for model compression and acceleration, both in academia and commercial AI chips. In retrospect, we review the background of this project, summarize the pros and cons, and discuss new opportunities where pruning, sparsity, and low precision can accelerate emerging deep learning workloads.

cs.AR↗
arXiv2023

FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention

Diffusion models excel at text-to-image generation, especially in subject-driven generation for personalized images. However, existing methods are inefficient due to the subject-specific fine-tuning, which is computationally intensive and hampers efficient deployment. Moreover, existing methods struggle with multi-subject generation as they often blend features among subjects. We present FastComposer which enables efficient, personalized, multi-subject text-to-image generation without fine-tuning. FastComposer uses subject embeddings extracted by an image encoder to augment the generic text conditioning in diffusion models, enabling personalized image generation based on subject images and textual instructions with only forward passes. To address the identity blending problem in the multi-subject generation, FastComposer proposes cross-attention localization supervision during training, enforcing the attention of reference subjects localized to the correct regions in the target images. Naively conditioning on subject embeddings results in subject overfitting. FastComposer proposes delayed subject conditioning in the denoising step to maintain both identity and editability in subject-driven image generation. FastComposer generates images of multiple unseen individuals with different styles, actions, and contexts. It achieves 300$\times$-2500$\times$ speedup compared to fine-tuning-based methods and requires zero extra storage for new subjects. FastComposer paves the way for efficient, personalized, and high-quality multi-subject image creation. Code, model, and dataset are available at https://github.com/mit-han-lab/fastcomposer.

cs.CV↗
arXiv2023

SparseViT: Revisiting Activation Sparsity for Efficient High-Resolution Vision Transformer

High-resolution images enable neural networks to learn richer visual representations. However, this improved performance comes at the cost of growing computational complexity, hindering their usage in latency-sensitive applications. As not all pixels are equal, skipping computations for less-important regions offers a simple and effective measure to reduce the computation. This, however, is hard to be translated into actual speedup for CNNs since it breaks the regularity of the dense convolution workload. In this paper, we introduce SparseViT that revisits activation sparsity for recent window-based vision transformers (ViTs). As window attentions are naturally batched over blocks, actual speedup with window activation pruning becomes possible: i.e., ~50% latency reduction with 60% sparsity. Different layers should be assigned with different pruning ratios due to their diverse sensitivities and computational costs. We introduce sparsity-aware adaptation and apply the evolutionary search to efficiently find the optimal layerwise sparsity configuration within the vast search space. SparseViT achieves speedups of 1.5x, 1.4x, and 1.3x compared to its dense counterpart in monocular 3D object detection, 2D instance segmentation, and 2D semantic segmentation, respectively, with negligible to no loss of accuracy.

cs.CV↗
arXiv2023

A Fast Algorithm for Source-wise Round-trip Spanners

In this paper, we study the problem of fast constructions of source-wise round-trip spanners in weighted directed graphs. For a source vertex set $S\subseteq V$ in a graph $G(V,E)$, an $S$-sourcewise round-trip spanner of $G$ of stretch $k$ is a subgraph $H$ of $G$ such that for every pair of vertices $u,v\in S\times V$, their round-trip distance in $H$ is at most $k$ times of their round-trip distance in $G$. We show that for a graph $G(V,E)$ with $n$ vertices and $m$ edges, an $s$-sized source vertex set $S\subseteq V$ and an integer $k>1$, there exists an algorithm that in time $O(ms^{1/k}\log^5n)$ constructs an $S$-sourcewise round-trip spanner of stretch $O(k\log n)$ and $O(ns^{1/k}\log^2n)$ edges with high probability. Compared to the fast algorithms for constructing all-pairs round-trip spanners \cite{PRS+18,CLR+20}, our algorithm improve the running time and the number of edges in the spanner when $k$ is super-constant. Compared with the existing algorithm for constructing source-wise round-trip spanners \cite{ZL17}, our algorithm significantly improves their construction time $Ω(\min\{ms,n^ω\})$ (where $ω\in [2,2.373)$ and 2.373 is the matrix multiplication exponent) to nearly linear $O(ms^{1/k}\log^5n)$, at the expense of paying an extra $O(\log n)$ in the stretch. As an important building block of the algorithm, we develop a graph partitioning algorithm to partition $G$ into clusters of bounded radius and prove that for every $u,v\in S\times V$ at small round-trip distance, the probability of separating them in different clusters is small. The algorithm takes the size of $S$ as input and does not need the knowledge of $S$. With the algorithm and a reachability vertex size estimation algorithm, we show that the recursive algorithm for constructing standard round-trip spanners \cite{PRS+18} can be adapted to the source-wise setting.

cs.DS↗
arXiv2023

Offsite-Tuning: Transfer Learning without Full Model

Transfer learning is important for foundation models to adapt to downstream tasks. However, many foundation models are proprietary, so users must share their data with model owners to fine-tune the models, which is costly and raise privacy concerns. Moreover, fine-tuning large foundation models is computation-intensive and impractical for most downstream users. In this paper, we propose Offsite-Tuning, a privacy-preserving and efficient transfer learning framework that can adapt billion-parameter foundation models to downstream data without access to the full model. In offsite-tuning, the model owner sends a light-weight adapter and a lossy compressed emulator to the data owner, who then fine-tunes the adapter on the downstream data with the emulator's assistance. The fine-tuned adapter is then returned to the model owner, who plugs it into the full model to create an adapted foundation model. Offsite-tuning preserves both parties' privacy and is computationally more efficient than the existing fine-tuning methods that require access to the full model weights. We demonstrate the effectiveness of offsite-tuning on various large language and vision foundation models. Offsite-tuning can achieve comparable accuracy as full model fine-tuning while being privacy-preserving and efficient, achieving 6.5x speedup and 5.6x memory reduction. Code is available at https://github.com/mit-han-lab/offsite-tuning.

cs.CL↗
arXiv2022

Hybrid Gate-Pulse Model for Variational Quantum Algorithms

Current quantum programs are mostly synthesized and compiled on the gate-level, where quantum circuits are composed of quantum gates. The gate-level workflow, however, introduces significant redundancy when quantum gates are eventually transformed into control signals and applied on quantum devices. For superconducting quantum computers, the control signals are microwave pulses. Therefore, pulse-level optimization has gained more attention from researchers due to their advantages in terms of circuit duration. Recent works, however, are limited by their poor scalability brought by the large parameter space of control signals. In addition, the lack of gate-level "knowledge" also affects the performance of pure pulse-level frameworks. We present a hybrid gate-pulse model that can mitigate these problems. We propose to use gate-level compilation and optimization for "fixed" part of the quantum circuits and to use pulse-level methods for problem-agnostic parts. Experimental results demonstrate the efficiency of the proposed framework in discrete optimization tasks. We achieve a performance boost at most 8% with 60% shorter pulse duration in the problem-agnostic layer.

quant-ph↗
arXiv2022

Composite Resource Scheduling for Networked Control Systems

Real-time end-to-end task scheduling in networked control systems (NCSs) requires the joint consideration of both network and computing resources to guarantee the desired quality of service (QoS). This paper introduces a new model for composite resource scheduling (CRS) in real-time networked control systems, which considers a strict execution order of sensing, computing, and actuating segments based on the control loop of the target NCS. We prove that the general CRS problem is NP-hard and study two special cases of the CRS problem. The first case restricts the computing and actuating segments to have unit-size execution time while the second case assumes that both sensing and actuating segments have unit-size execution time. We propose an optimal algorithm to solve the first case by checking the intervals with 100% network resource utilization and modify the deadlines of the tasks within those intervals to prune the search. For the second case, we propose another optimal algorithm based on a novel backtracking strategy to check the time intervals with the network resource utilization larger than 100% and modify the timing parameters of tasks based on these intervals. For the general case, we design a greedy strategy to modify the timing parameters of both network segments and computing segments within the time intervals that have network and computing resource utilization larger than 100%, respectively. The correctness and effectiveness of the proposed algorithms are verified through extensive experiments.

cs.NI↗
arXiv2022

TopGen: Topology-Aware Bottom-Up Generator for Variational Quantum Circuits

Variational Quantum Algorithms (VQA) are promising to demonstrate quantum advantages on near-term devices. Designing ansatz, a variational circuit with parameterized gates, is of paramount importance for VQA as it lays the foundation for parameter optimizations. Due to the large noise on Noisy-Intermediate Scale Quantum (NISQ) machines, considering circuit size and real device noise in the ansatz design process is necessary. Unfortunately, recent works on ansatz design either consider no noise impact or only treat the real device as a black box with no specific noise information. In this work, we propose to open the black box by designing specific ansatz tailored for the qubit topology on target machines. Specifically, we propose a bottom-up approach to generate topology-specific ansatz. Firstly, we generate topology-compatible sub-circuits with desirable properties such as high expressibility and entangling capability. Then, the sub-circuits are combined together to form an initial ansatz. We further propose circuits stitching to solve the sparse connectivity issue between sub-circuits, and dynamic circuit growing to improve the accuracy. The ansatz constructed with this method is highly flexible and thus we can explore a much larger design space than previous state-of-the-art method in which all ansatz candidates are strict subsets of a pre-defined large ansatz. We use a popular VQA algorithm - Quantum Neural Networks (QNN) for Machine Learning (ML) task as the benchmarks. Experiments on 14 ML tasks show that under the same performance, the TopGen-searched ansatz can reduce the circuit depth and the number of CNOT gates by up to 2 * and 4 * respectively. Experiments on three real quantum machines demonstrate on average 17% accuracy improvements over baselines.

quant-ph↗
arXiv2022

Demonstration of broadband topological slow light

Slow-light devices are able to significantly enhance light-matter interaction due to the reduced group velocity of light, but a very low group velocity is usually achieved in a narrow bandwidth, accompanied by extreme sensitivity to imperfections that causes increased disorder-induced attenuation. Recent theories have suggested an ideal solution to this problem - unidirectional chiral photonic states, previously discovered in structures known as photonic topological insulators, not only resist backscattering from imperfections but can also be slowed down in the entire topological bandgap with multiple windings in the Brillouin zone. Here, we report on the experimental demonstration of broadband topological slow light in a photonic topological insulator. When coupled with periodic resonators that form flat bands, the chiral photonic states can wind many times around the Brillouin zone, achieving an ultra-low group velocity in the entire topological bandgap. This demonstration extends the scope of topological photonics into slow light engineering and opens a unique avenue in the dispersion manipulation of chiral photonic states.

physics.app-ph↗
arXiv2022

Variational Quantum Pulse Learning

Quantum computing is among the most promising emerging techniques to solve problems that are computationally intractable on classical hardware. A large body of existing works focus on using variational quantum algorithms on the gate level for machine learning tasks, such as the variational quantum circuit (VQC). However, VQC has limited flexibility and expressibility due to limited number of parameters, e.g. only one parameter can be trained in one rotation gate. On the other hand, we observe that quantum pulses are lower than quantum gates in the stack of quantum computing and offers more control parameters. Inspired by the promising performance of VQC, in this paper we propose variational quantum pulses (VQP), a novel paradigm to directly train quantum pulses for learning tasks. The proposed method manipulates variational quantum pulses by pulling and pushing the amplitudes of pulses in an optimization framework. Similar to variational quantum algorithms, our framework to train pulses maintains the robustness to noise on Noisy Intermediate-Scale Quantum (NISQ) computers. In an example task of binary classification, VQP learning achieves up to 11% and 9% higher accuracy compared with VQC learning on the qiskit noise simulators (with noise model from real machine) and ibmq-jarkata, respectively, demonstrating its effectiveness and feasibility. Stability for VQP to obtain reliable results has also been verified in the presence of noise.

quant-ph↗