arXiv subjects

Zhewen Yu

Publications and source records attributed to Zhewen Yu.

12 recordsLinked to original sources

arXiv2025

From Loop Nests to Silicon: Mapping AI Workloads onto AMD NPUs with MLIR-AIR

General-purpose compilers abstract away parallelism, locality, and synchronization, limiting their effectiveness on modern spatial architectures. As modern computing architectures increasingly rely on fine-grained control over data movement, execution order, and compute placement for performance, compiler infrastructure must provide explicit mechanisms for orchestrating compute and data to fully exploit such architectures. We introduce MLIR-AIR, a novel, open-source compiler stack built on MLIR that bridges the semantic gap between high-level workloads and fine-grained spatial architectures such as AMD's NPUs. MLIR-AIR defines the AIR dialect, which provides structured representations for asynchronous and hierarchical operations across compute and memory resources. AIR primitives allow the compiler to orchestrate spatial scheduling, distribute computation across hardware regions, and overlap communication with computation without relying on ad hoc runtime coordination or manual scheduling. We demonstrate MLIR-AIR's capabilities through two case studies: matrix multiplication and the multi-head attention block from the LLaMA 2 model. For matrix multiplication, MLIR-AIR achieves up to 78.7% compute efficiency and generates implementations with performance almost identical to state-of-the-art, hand-optimized matrix multiplication written using the lower-level, close-to-metal MLIR-AIE framework. For multi-head attention, we demonstrate that the AIR interface supports fused implementations using approximately 150 lines of code, enabling tractable expression of complex workloads with efficient mapping to spatial hardware. MLIR-AIR transforms high-level structured control flow into spatial programs that efficiently utilize the compute fabric and memory hierarchy of an NPU, leveraging asynchronous execution, tiling, and communication overlap through compiler-managed scheduling.

cs.CL↗

arXiv2025

ITERA-LLM: Boosting Sub-8-Bit Large Language Model Inference via Iterative Tensor Decomposition

Recent advancements in Large Language Models (LLMs) have demonstrated impressive capabilities as their scale expands to billions of parameters. Deploying these large-scale models on resource-constrained platforms presents significant challenges, with post-training fixed-point quantization often used as a model compression technique. However, quantization-only methods typically lead to significant accuracy degradation in LLMs when precision falls below 8 bits. This paper addresses this challenge through a software-hardware co-design framework, ITERA-LLM, which integrates sub-8-bit quantization with SVD-based iterative low-rank tensor decomposition for error compensation, leading to higher compression ratios and reduced computational complexity. The proposed approach is complemented by a hardware-aware Design Space Exploration (DSE) process that optimizes accuracy, latency, and resource utilization, tailoring the configuration to the specific requirements of the targeted LLM. Our results show that ITERA-LLM achieves linear layer latency reduction of up to 41.1%, compared to quantization-only baseline approach while maintaining similar model accuracy.

cs.AR↗

arXiv2024

Align$^2$LLaVA: Cascaded Human and Large Language Model Preference Alignment for Multi-modal Instruction Curation

Recent advances in Multi-modal Large Language Models (MLLMs), such as LLaVA-series models, are driven by massive machine-generated instruction-following data tuning. Such automatic instruction collection pipelines, however, inadvertently introduce significant variability in data quality. This paper introduces a novel instruction curation algorithm, derived from two unique perspectives, human and LLM preference alignment, to compress this vast corpus of machine-generated multimodal instructions to a compact and high-quality form: (i) For human preference alignment, we have collected a machine-generated multimodal instruction dataset and established a comprehensive set of both subjective and objective criteria to guide the data quality assessment critically from human experts. By doing so, a reward model was trained on the annotated dataset to internalize the nuanced human understanding of instruction alignment. (ii) For LLM preference alignment, given the instruction selected by the reward model, we propose leveraging the inner LLM used in MLLM to align the writing style of visual instructions with that of the inner LLM itself, resulting in LLM-aligned instruction improvement. Extensive experiments demonstrate that we can maintain or even improve model performance by compressing synthetic multimodal instructions by up to 90%. Impressively, by aggressively reducing the training instructions from 158k to 14k (9$\times$ smaller), our model consistently outperforms its full-size dataset counterpart across various MLLM benchmarks. Our project is available at https://github.com/DCDmllm/Align2LLaVA.

cs.AI↗

arXiv2024

HASS: Hardware-Aware Sparsity Search for Dataflow DNN Accelerator

Deep Neural Networks (DNNs) excel in learning hierarchical representations from raw data, such as images, audio, and text. To compute these DNN models with high performance and energy efficiency, these models are usually deployed onto customized hardware accelerators. Among various accelerator designs, dataflow architecture has shown promising performance due to its layer-pipelined structure and its scalability in data parallelism. Exploiting weights and activations sparsity can further enhance memory storage and computation efficiency. However, existing approaches focus on exploiting sparsity in non-dataflow accelerators, which cannot be applied onto dataflow accelerators because of the large hardware design space introduced. As such, this could miss opportunities to find an optimal combination of sparsity features and hardware designs. In this paper, we propose a novel approach to exploit unstructured weights and activations sparsity for dataflow accelerators, using software and hardware co-optimization. We propose a Hardware-Aware Sparsity Search (HASS) to systematically determine an efficient sparsity solution for dataflow accelerators. Over a set of models, we achieve an efficiency improvement ranging from 1.3$\times$ to 4.2$\times$ compared to existing sparse designs, which are either non-dataflow or non-hardware-aware. Particularly, the throughput of MobileNetV3 can be optimized to 4895 images per second. HASS is open-source: \url{https://github.com/Yu-Zhewen/HASS}

cs.AR↗

arXiv2024

SMOF: Streaming Modern CNNs on FPGAs with Smart Off-Chip Eviction

Convolutional Neural Networks (CNNs) have demonstrated their effectiveness in numerous vision tasks. However, their high processing requirements necessitate efficient hardware acceleration to meet the application's performance targets. In the space of FPGAs, streaming-based dataflow architectures are often adopted by users, as significant performance gains can be achieved through layer-wise pipelining and reduced off-chip memory access by retaining data on-chip. However, modern topologies, such as the UNet, YOLO, and X3D models, utilise long skip connections, requiring significant on-chip storage and thus limiting the performance achieved by such system architectures. The paper addresses the above limitation by introducing weight and activation eviction mechanisms to off-chip memory along the computational pipeline, taking into account the available compute and memory resources. The proposed mechanism is incorporated into an existing toolflow, expanding the design space by utilising off-chip memory as a buffer. This enables the mapping of such modern CNNs to devices with limited on-chip memory, under the streaming architecture design approach. SMOF has demonstrated the capacity to deliver competitive and, in some cases, state-of-the-art performance across a spectrum of computer vision tasks, achieving up to 10.65 X throughput improvement compared to previous works.

cs.AR↗

arXiv2023

AutoWS: Automate Weights Streaming in Layer-wise Pipelined DNN Accelerators

With the great success of Deep Neural Networks (DNN), the design of efficient hardware accelerators has triggered wide interest in the research community. Existing research explores two architectural strategies: sequential layer execution and layer-wise pipelining. While the former supports a wider range of models, the latter is favoured for its enhanced customization and efficiency. A challenge for the layer-wise pipelining architecture is its substantial demand for the on-chip memory for weights storage, impeding the deployment of large-scale networks on resource-constrained devices. This paper introduces AutoWS, a pioneering memory management methodology that exploits both on-chip and off-chip memory to optimize weight storage within a layer-wise pipelining architecture, taking advantage of its static schedule. Through a comprehensive investigation on both the hardware design and the Design Space Exploration, our methodology is fully automated and enables the deployment of large-scale DNN models on resource-constrained devices, which was not possible in existing works that target layer-wise pipelining architectures. AutoWS is open-source: https://github.com/Yu-Zhewen/AutoWS

cs.AR↗

arXiv2023

SATAY: A Streaming Architecture Toolflow for Accelerating YOLO Models on FPGA Devices

AI has led to significant advancements in computer vision and image processing tasks, enabling a wide range of applications in real-life scenarios, from autonomous vehicles to medical imaging. Many of those applications require efficient object detection algorithms and complementary real-time, low latency hardware to perform inference of these algorithms. The YOLO family of models is considered the most efficient for object detection, having only a single model pass. Despite this, the complexity and size of YOLO models can be too computationally demanding for current edge-based platforms. To address this, we present SATAY: a Streaming Architecture Toolflow for Accelerating YOLO. This work tackles the challenges of deploying stateof-the-art object detection models onto FPGA devices for ultralow latency applications, enabling real-time, edge-based object detection. We employ a streaming architecture design for our YOLO accelerators, implementing the complete model on-chip in a deeply pipelined fashion. These accelerators are generated using an automated toolflow, and can target a range of suitable FPGA devices. We introduce novel hardware components to support the operations of YOLO models in a dataflow manner, and off-chip memory buffering to address the limited on-chip memory resources. Our toolflow is able to generate accelerator designs which demonstrate competitive performance and energy characteristics to GPU devices, and which outperform current state-of-the-art FPGA accelerators.

cs.AR↗

arXiv2023

A Dataflow Compiler for Efficient LLM Inference using Custom Microscaling Formats

Model quantization represents both parameters (weights) and intermediate values (activations) in a more compact format, thereby directly reducing both computational and memory cost in hardware. The quantization of recent large language models (LLMs) faces challenges to achieve competitive memory density compared to other models such as convolutional neural networks, since values in LLMs require larger dynamic ranges. Current hardware can expedite computation for LLMs using compact numerical formats such as low-bitwidth integers or floating-point numbers. Each has advantages: integer operations simplify circuit design, whereas floating-point calculations can enhance accuracy when a wider dynamic range is required. In this work, we seek an efficient data format that combines the best of both worlds: Microscaling (MX) formats. MX formats are efficient data formats that achieve both large dynamic ranges and high memory density. In this paper, we propose a compiler named MASE for exploring mixed-precision MX formats on dataflow hardware accelerators for LLM inference. Our main contributions are twofold. First, we propose a novel orchestration abstraction to explore both software and hardware optimizations with new data formats. Second, MASE achieves LLM inference at an average precision of 4-bits, with minimal to no accuracy degradation. To our knowledge, MASE represents the first effort to harness fine-grain multi-precision MX formats in the design of LLM hardware accelerators. Over a range of LLMs and datasets, MASE achieves an average improvement of 24% in $\Delta$ accuracy with an overhead of only 3% in energy efficiency compared to designs using 8-bit fixed-point numbers.

cs.AR↗

arXiv2023

PASS: Exploiting Post-Activation Sparsity in Streaming Architectures for CNN Acceleration

With the ever-growing popularity of Artificial Intelligence, there is an increasing demand for more performant and efficient underlying hardware. Convolutional Neural Networks (CNN) are a workload of particular importance, which achieve high accuracy in computer vision applications. Inside CNNs, a significant number of the post-activation values are zero, resulting in many redundant computations. Recent works have explored this post-activation sparsity on instruction-based CNN accelerators but not on streaming CNN accelerators, despite the fact that streaming architectures are considered the leading design methodology in terms of performance. In this paper, we highlight the challenges associated with exploiting post-activation sparsity for performance gains in streaming CNN accelerators, and demonstrate our approach to address them. Using a set of modern CNN benchmarks, our streaming sparse accelerators achieve 1.41x to 1.93x efficiency (GOP/s/DSP) compared to state-of-the-art instruction-based sparse accelerators.

cs.AR↗

arXiv2023

Mixed-TD: Efficient Neural Network Accelerator with Layer-Specific Tensor Decomposition

Neural Network designs are quite diverse, from VGG-style to ResNet-style, and from Convolutional Neural Networks to Transformers. Towards the design of efficient accelerators, many works have adopted a dataflow-based, inter-layer pipelined architecture, with a customised hardware towards each layer, achieving ultra high throughput and low latency. The deployment of neural networks to such dataflow architecture accelerators is usually hindered by the available on-chip memory as it is desirable to preload the weights of neural networks on-chip to maximise the system performance. To address this, networks are usually compressed before the deployment through methods such as pruning, quantization and tensor decomposition. In this paper, a framework for mapping CNNs onto FPGAs based on a novel tensor decomposition method called Mixed-TD is proposed. The proposed method applies layer-specific Singular Value Decomposition (SVD) and Canonical Polyadic Decomposition (CPD) in a mixed manner, achieving 1.73x to 10.29x throughput per DSP to state-of-the-art CNNs. Our work is open-sourced: https://github.com/Yu-Zhewen/Mixed-TD

cs.LG↗

arXiv2022

SVD-NAS: Coupling Low-Rank Approximation and Neural Architecture Search

The task of compressing pre-trained Deep Neural Networks has attracted wide interest of the research community due to its great benefits in freeing practitioners from data access requirements. In this domain, low-rank approximation is a promising method, but existing solutions considered a restricted number of design choices and failed to efficiently explore the design space, which lead to severe accuracy degradation and limited compression ratio achieved. To address the above limitations, this work proposes the SVD-NAS framework that couples the domains of low-rank approximation and neural architecture search. SVD-NAS generalises and expands the design choices of previous works by introducing the Low-Rank architecture space, LR-space, which is a more fine-grained design space of low-rank approximation. Afterwards, this work proposes a gradient-descent-based search for efficiently traversing the LR-space. This finer and more thorough exploration of the possible design choices results in improved accuracy as well as reduction in parameters, FLOPS, and latency of a CNN model. Results demonstrate that the SVD-NAS achieves 2.06-12.85pp higher accuracy on ImageNet than state-of-the-art methods under the data-limited problem setting. SVD-NAS is open-sourced at https://github.com/Yu-Zhewen/SVD-NAS.

cs.LG↗

arXiv2021

SAMO: Optimised Mapping of Convolutional Neural Networks to Streaming Architectures

Significant effort has been placed on the development of toolflows that map Convolutional Neural Network (CNN) models to Field Programmable Gate Arrays (FPGAs) with the aim of automating the production of high performing designs for a diverse set of applications. However, within these toolflows, the problem of finding an optimal mapping is often overlooked, with the expectation that the end user will tune their generated hardware for their desired platform. This is particularly prominent within Streaming Architecture toolflows, where there is a large design space to explore . In this work, we establish the framework SAMO: a Streaming Architecture Mapping Optimiser. SAMO exploits the structure of CNN models and the common features that exist in Streaming Architectures, and casts the mapping optimisation problem under a unified methodology. Furthermore, SAMO explicitly explores the reconfigurability property of FPGAs, allowing the methodology to overcome mapping limitations imposed by certain toolflows under resource-constrained scenarios, as well as improve on the achievable throughput. Three optimisation methods - Brute-Force, Simulated Annealing and Rule-Based - have been developed in order to generate valid, high performance designs for a range of target platforms and CNN models. Results show that SAMO-optimised designs can achieve 4-20x better performance compared to existing hand-tuned designs. The SAMO framework is open-source: https://github.com/AlexMontgomerie/samo.

cs.AR↗