FP8 low-precision formats have gained significant adoption in transformer inference and training. However, existing digital compute-in-memory (DCIM) architectures face challenges in supporting variable FP8 aligned-mantissa bitwidths, as unified alignment strategies and fixed-precision multiply accumulate (MAC) units struggle to handle input data with diverse distributions. This work presents a flexible FP8 DCIM accelerator with three innovations: 1) a dynamic shift-aware bitwidth prediction (DSBP) with on-the-fly input prediction that adaptively adjusts weight (2/4/6/8b) and input ($2\sim 12$b) aligned-mantissa precision; 2) a FIFO-based input alignment unit (FIAU) replacing complex barrel shifters with pointer-based control; and 3) a precision-scalable INT MAC array achieving flexible weight precision with minimal overhead. Implemented in 28-nm CMOS with a$64~\times ~96$CIM array, the design achieves 20.4 TFLOPS/W for fixed E5M7, demonstrating$2.8\times $higher FP8 efficiency than previous work while supporting all FP8 formats. Results on Llama-7b show that the DSBP achieves higher efficiency than fixed bitwidth mode at the same accuracy level on both BoolQ and Winogrande datasets, with configurable parameters enabling flexible accuracy–efficiency tradeoffs.

Text-to-Image (T2I) diffusion models have achieved remarkable success in image generation. Despite their progress, challenges remain in both prompt-following ability, image quality and lack of high-quality datasets, which are essential for refining these models. As acquiring labeled data is costly, we introduce AGFSync, a framework that enhances T2I diffusion models through Direct Preference Optimization (DPO) in a fully AI-driven approach. AGFSync utilizes Vision-Language Models (VLM) to assess image quality across style, coherence, and aesthetics, generating feedback data within an AI-driven loop. By applying AGFSync to leading T2I models such as SD v1.4, v1.5, and SDXL-base, our extensive experiments on the TIFA dataset demonstrate notable improvements in VQA scores, aesthetic evaluations, and performance on the HPS v2 benchmark, consistently outperforming the base models. AGFSync's method of refining T2I diffusion models paves the way for scalable alignment techniques.

Security concerns related to Large Language Models (LLMs) have been extensively explored; however, the safety implications for Multimodal Large Language Models (MLLMs), particularly in medical contexts (MedMLLMs), remain inadequately addressed. This paper investigates the security vulnerabilities of MedMLLMs, focusing on their deployment in clinical environments where the accuracy and relevance of question-and-answer interactions are crucial for addressing complex medical challenges. We introduce and redefine two attack types: mismatched malicious attack (2M-attack) and optimized mismatched malicious attack (O2M-attack), by integrating existing clinical data with atypical natural phenomena. Using the comprehensive 3MAD dataset that we developed, which spans a diverse range of medical imaging modalities and adverse medical scenarios, we performed an in-depth analysis and proposed the MCM optimization method. This approach significantly improves the attack success rate against MedMLLMs. Our evaluations, which include white-box attacks on LLaVA-Med and transfer (black-box) attacks on four other SOTA models, reveal that even MedMLLMs designed with advanced security mechanisms remain vulnerable to breaches. This study highlights the critical need for robust security measures to enhance the safety and reliability of open-source MedMLLMs, especially in light of the potential impact of jailbreak attacks and other malicious exploits in clinical applications. Warning: Medical jailbreaking may generate content that includes unverified diagnoses and treatment recommendations. Always consult professional medical advice.

Existing text-to-image (T2I) diffusion models face several limitations, including large model sizes, slow runtime, and low-quality generation on mobile devices. This paper aims to address all of these challenges by developing an extremely small and fast T2I model that generates high-resolution and high-quality images on mobile platforms. We propose several techniques to achieve this goal. First, we systematically examine the design choices of the network architecture to reduce model parameters and latency, while ensuring high-quality generation. Second, to further improve generation quality, we employ cross-architecture knowledge distillation from a much larger model, using a multi-level approach to guide the training of our model from scratch. Third, we enable a few-step generation by integrating adversarial guidance with knowledge distillation. For the first time, our model SnapGen, demonstrates the generation of 10242px images on a mobile device around 1.4 seconds. On ImageNet-1K, our model, with only 372M parameters, achieves an FID of 2.06 for 2562px generation. On T2I benchmarks (i.e., GenEval and DPG-Bench), our model with merely 379M parameters, surpasses large-scale models with billions of parameters at a significantly smaller size (e.g., 7× smaller than SDXL, 14× smaller than IF-XL).

DNN accelerators, significantly advanced by model compression and specialized dataflow techniques, have marked considerable progress. However, the frequent access of highprecision partial sums (PSUMs) leads to excessive memory demands in architectures utilizing input/weight stationary dataflows. Traditional compression strategies have typically overlooked PSUM quantization, which may account for 69% of power consumption. This study introduces a novel Additive Partial Sum Quantization (APSQ) method, seamlessly integrating PSUM accumulation into the quantization framework. A grouping strategy that combines APSQ with PSUM quantization enhanced by a reconfigurable architecture is further proposed. The APSQ performs nearly lossless on NLP and CV tasks across BERT, Segformer, and EfficientViT models while compressing PSUMs to INT8. This leads to a notable reduction in energy costs by $\mathbf{2 8-8 7 \%}$. Extended experiments on LLaMA2-7B demonstrate the potential of APSQ for large language models. Code is available at https://github.com/Yonghao-Tan/APSQ.

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by integrating external knowledge retrieval but faces challenges on edge devices due to high storage, energy, and latency demands. Computing-in-Memory (CIM) offers a promising solution by storing document embeddings in CIM macros and enabling in-situ parallel retrievals but is constrained by either low memory density or limited computational accuracy. To address these challenges, we present DIRC-RAG, a novel edge RAG acceleration architecture leveraging Digital In-ReRAM Computation (DIRC). DIRC integrates a high-density multi-level ReRAM subarray with an SRAM cell, utilizing SRAM and differential sensing for robust ReRAM readout and digital multiply-accumulate (MAC) operations. By storing all document embeddings within the CIM macro, DIRC achieves ultra-low-power, single-cycle data loading, substantially reducing both energy consumption and latency compared to off-chip DRAM. A query-stationary (QS) dataflow is supported for RAG tasks, minimizing on-chip data movement and reducing SRAM buffer requirements. We introduce error optimization for the DIRC ReRAM-SRAM cell by extracting the bit-wise spatial error distribution of the ReRAM subarray and applying targeted bit-wise data remapping. An error detection circuit is also implemented to enhance readout resilience against device-and circuit-level variations.Simulation results demonstrate that DIRC-RAG under TSMC 40nm process achieves an on-chip non-volatile memory density of 5.18Mb/mm2and a throughput of 131 TOPS. It delivers a 4MB retrieval latency of 5.6μs/query and an energy consumption of 0.956μJ/query, while maintaining the retrieval precision.

Non-linear functions are prevalent in Transformers and their lightweight variants, incurring substantial and frequently underestimated hardware costs. Previous state-of-the-art works optimize these operations by piece-wise linear approximation and store the parameters in look-up tables (LUT), but most of them require unfriendly high-precision arithmetics such as FP/INT 32 and lack consideration of integer-only INT quantization. This paper proposed a genetic LUT-Approximation algorithm namely GQA-LUT that can automatically determine the parameters with quantization awareness. The results demonstrate that GQA-LUT achieves negligible degradation on the challenging semantic segmentation task for both vanilla and linear Transformer models. Besides, proposed GQA-LUT enables the employment of INT8-based LUT-Approximation that achieves an area savings of 81.3~81.7% and a power reduction of 79.3~80.2% compared to the high-precision FP/INT 32 alternatives. Code is available at https://github.com/PingchengDong/GQA-LUT.

Large Language Models (LLMs) have shown impressive capabilities, yet they still struggle with math reasoning.In this work, we propose CoT-Influx, a novel approach that pushes the boundary of few-shot Chain-of-Thoughts (CoT) learning to improve LLM mathematical reasoning.Motivated by the observation that adding more concise CoT examples in the prompt can improve LLM reasoning performance, CoT-Influx employs a coarse-to-fine pruner to maximize the input of effective and concise CoT examples.The pruner first selects as many crucial CoT examples as possible and then prunes unimportant tokens to fit the context window.A math reasoning dataset with diverse difficulty levels and reasoning steps is used to train the pruner, along with a math-specialized reinforcement learning approach.As a result, by enabling more CoT examples with double the context window size in tokens, CoT-Influx significantly outperforms various prompting baselines across various LLMs (LLaMA2-7B, 13B, 70B) and 6 math datasets, achieving up to 4.40% absolute improvements.Remarkably, without any fine-tuning, LLaMA2-70B with CoT-Influx surpasses GPT-3.5 and a wide range of larger LLMs (PaLM, Minerva 540B, etc.) on GSM8K.CoT-Influx is a plug-and-play module for LLMs, adaptable in various scenarios.It's compatible with advanced reasoning prompting techniques, such as self-consistency, and supports different long-context LLMs, including Mistral-7B-v0.3-32K and Yi-6B-200K.Codes are available at https://github.com/HuangOwen/CoT-Influx

Constructing supervised machine learning models for real-world video analysis require substantial labeled data, which is costly to acquire due to scarce domain expertise and laborious manual inspection. While data programming shows promise in generating labeled data at scale with user-defined labeling functions, the high dimensional and complex temporal information in videos poses additional challenges for effectively composing and evaluating labeling functions. In this paper, we propose VideoPro, a visual analytics approach to support flexible and scalable video data programming for model steering with reduced human effort. We first extract human-understandable events from videos using computer vision techniques and treat them as atomic components of labeling functions. We further propose a two-stage template mining algorithm that characterizes the sequential patterns of these events to serve as labeling function templates for efficient data labeling. The visual interface of VideoPro facilitates multifaceted exploration, examination, and application of the labeling templates, allowing for effective programming of video data at scale. Moreover, users can monitor the impact of programming on model performance and make informed adjustments during the iterative programming process. We demonstrate the efficiency and effectiveness of our approach with two case studies and expert interviews.

We propose LLM-FP4 for quantizing both weights and activations in large language models (LLMs) down to 4-bit floating-point values, in a post-training manner.Existing posttraining quantization (PTQ) solutions are primarily integer-based and struggle with bit widths below 8 bits.Compared to integer quantization, floating-point (FP) quantization is more flexible and can better handle long-tail or bell-shaped distributions, and it has emerged as a default choice in many hardware platforms.One characteristic of FP quantization is that its performance largely depends on the choice of exponent bits and clipping range.In this regard, we construct a strong FP-PTQ baseline by searching for the optimal quantization parameters.Furthermore, we observe a high interchannel variance and low intra-channel variance pattern in activation distributions, which adds activation quantization difficulty.We recognize this pattern to be consistent across a spectrum of transformer models designed for diverse tasks, such as LLMs, BERT, and Vision Transformer models.To tackle this, we propose per-channel activation quantization and show that these additional scaling factors can be reparameterized as exponential biases of weights, incurring a negligible cost.Our method, for the first time, can quantize both weights and activations in the LLaMA-13B to only 4-bit and achieves an average score of 63.1 on the common sense zero-shot reasoning tasks, which is only 5.8 lower than the full-precision model, significantly outperforming the previous stateof-the-art by 12.7 points.Code is available at: https://github.com/nbasyl/LLM-FP4.

The purpose of federated learning is to enable multiple clients to jointly train a machine learning model without sharing data. However, the existing methods for training an image segmentation model have been based on an unrealistic assumption that the training set for each local client is annotated in a similar fashion and thus follows the same image supervision level. To relax this assumption, in this work, we propose a label-agnostic unified federated learning framework, named FedMix, for medical image segmentation based on mixed image labels. In FedMix, each client updates the federated model by integrating and effectively making use of all available labeled data ranging from strong pixel-level labels, weak bounding box labels, to weakest image-level class labels. Based on these local models, we further propose an adaptive weight assignment procedure across local clients, where each client learns an aggregation weight during the global model update. Compared to the existing methods, FedMix not only breaks through the constraint of a single level of image supervision but also can dynamically adjust the aggregation weight of each local client, achieving rich yet discriminative feature representations. Experimental results on multiple publicly-available datasets validate that the proposed FedMix outperforms the state-of-the-art methods by a large margin. In addition, we demonstrate through experiments that FedMix is extendable to multi-class medical image segmentation and much more feasible in clinical scenarios. The code is available at: https://github.com/Jwicaksana/FedMix.

In order to deploy deep models in a computationally efficient manner, model quantization approaches have been frequently used. In addition, as new hardware that supports various-bit arithmetic operations, recent research on mixed precision quantization (MPQ) begins to fully leverage the capacity of representation by searching various bitwidths for different layers and modules in a network. However, previous studies mainly search the MPQ strategy in a costly scheme using reinforcement learning, neural architecture search, etc., or simply utilize partial prior knowledge for bitwidth distribution, which might be biased and sub-optimal. In this work, we present a novel Stochastic Differentiable Quantization (SDQ) method that can automatically learn the MPQ strategy in a more flexible and globally-optimized space with a smoother gradient approximation. Particularly, Differentiable Bitwidth Parameters (DBPs) are employed as the probability factors in stochastic quantization between adjacent bitwidth. After the optimal MPQ strategy is acquired, we further train our network with the entropy-aware bin regularization and knowledge distillation. We extensively evaluate our method on different networks, hardwares (GPUs and FPGA), and datasets. SDQ outperforms all other state-of-the-art mixed or single precision quantization with less bitwidth, and are even better than the original full-precision counterparts across various ResNet and MobileNet families, demonstrating the effectiveness and superiority of our method. Code will be publicly available.

Human-object interaction (HOI) Detection is an important problem to understand how humans interact with objects. In this paper, we explore Interactiveness Knowledge which indicates whether human and object interact with each other or not. We found that interactiveness knowledge can be learned across HOI datasets and alleviate the gap between diverse HOI category settings. Our core idea is to exploit an Interactiveness Network to learn the general interactiveness knowledge from multiple HOI datasets and perform Non-Interaction Suppression before HOI classification in inference. On account of the generalization of interactiveness, interactiveness network is a transferable knowledge learner and can be cooperated with any HOI detection models to achieve desirable results. We utilize the human instance and body part features together to learn the interactiveness in hierarchical paradigm, i.e., instance-level and body part-level interactivenesses. Thereafter, a consistency task is proposed to guide the learning and extract deeper interactive visual clues. We extensively evaluate the proposed method on HICO-DET, V-COCO, and a newly constructed HAKE-HOI dataset. With the learned interactiveness, our method outperforms state-of-the-art HOI detection methods, verifying its efficacy and flexibility. Code is available at https://github.com/DirtyHarryLYL/Transferable-Interactiveness-Network.

Existing image-based activity understanding methods mainly adopt direct mapping, i.e. from image to activity concepts, which may encounter performance bottleneck since the huge gap. In light of this, we propose a new path: infer human part states first and then reason out the activities based on part-level semantics. Human Body Part States (PaSta) are fine-grained action semantic tokens, e.g., which can compose the activities and help us step toward human activity knowledge engine. To fully utilize the power of PaSta, we build a large-scale knowledge base PaStaNet, which contains 7M+ PaSta annotations. And two corresponding models are proposed: first, we design a model named Activity2Vec to extract PaSta features, which aim to be general representations for various activities. Second, we use a PaSta-based Reasoning method to infer activities. Promoted by PaStaNet, our method achieves significant improvements, e.g. 6.4 and 13.9 mAP on full and one-shot sets of HICO in supervised learning, and 3.2 and 4.2 mAP on V-COCO and images-based AVA in transfer learning. Code and data are available at http://hake-mvig.cn/.

Human-Object Interaction (HOI) Detection is an important problem to understand how humans interact with objects. In this paper, we explore Interactiveness Knowledge which indicates whether human and object interact with each other or not. We found that interactiveness knowledge can be learned across HOI datasets, regardless of HOI category settings. Our core idea is to exploit an Interactiveness Network to learn the general interactiveness knowledge from multiple HOI datasets and perform Non-Interaction Suppression before HOI classification in inference. On account of the generalization of interactiveness, interactiveness network is a transferable knowledge learner and can be cooperated with any HOI detection models to achieve desirable results. We extensively evaluate the proposed method on HICO-DET and V-COCO datasets. Our framework outperforms state-of-the-art HOI detection results by a great margin, verifying its efficacy and flexibility. Code is available at https://github.com/DirtyHarryLYL/Transferable-Interactiveness-Network.