Pie: Pooling CPU Memory for LLM Inference

The rapid growth of LLMs has revolutionized natural language processing and AI analysis, but their increasing size and memory demands present significant challenges. A common solution is to spill over to CPU memory; however, traditional GPU-CPU memory swapping often results in higher latency and lower throughput. This paper introduces Pie, an LLM inference framework that addresses these challenges with performance-transparent swapping and adaptive expansion. By leveraging predictable memory access patterns and the high bandwidth of modern hardware like the NVIDIA GH200 Grace Hopper Superchip, Pie enables concurrent data swapping without affecting foreground computation, expanding effective memory without added latency. Adaptive expansion dynamically adjusts CPU memory allocation based on real-time information, optimizing memory usage and performance under varying conditions. Pie maintains low computation latency, high throughput, and high elasticity. Our experimental evaluation demonstrates that Pie achieves optimal swapping policy during cache warmup and effectively balances increased memory capacity with negligible impact on computation. With its extended capacity, Pie outperforms vLLM by up to 1.9X in throughput and 2X in latency. Additionally, Pie can reduce GPU memory usage by up to 1.67X while maintaining the same performance. Compared to FlexGen, an offline profiling-based swapping solution, Pie achieves magnitudes lower latency and 9.4X higher throughput.

Cross-Fundus Transformer for Multi-modal Diabetic Retinopathy Grading with Cataract

Diabetic retinopathy (DR) is a leading cause of blindness worldwide and a common complication of diabetes. As two different imaging tools for DR grading, color fundus photography (CFP) and infrared fundus photography (IFP) are highly-correlated and complementary in clinical applications. To the best of our knowledge, this is the first study that explores a novel multi-modal deep learning framework to fuse the information from CFP and IFP towards more accurate DR grading. Specifically, we construct a dual-stream architecture Cross-Fundus Transformer (CFT) to fuse the ViT-based features of two fundus image modalities. In particular, a meticulously engineered Cross-Fundus Attention (CFA) module is introduced to capture the correspondence between CFP and IFP images. Moreover, we adopt both the single-modality and multi-modality supervisions to maximize the overall performance for DR grading. Extensive experiments on a clinical dataset consisting of 1,713 pairs of multi-modal fundus images demonstrate the superiority of our proposed method. Our code will be released for public access.

Shopping MMLU: A Massive Multi-Task Online Shopping Benchmark for Large Language Models

Online shopping is a complex multi-task, few-shot learning problem with a wide and evolving range of entities, relations, and tasks. However, existing models and benchmarks are commonly tailored to specific tasks, falling short of capturing the full complexity of online shopping. Large Language Models (LLMs), with their multi-task and few-shot learning abilities, have the potential to profoundly transform online shopping by alleviating task-specific engineering efforts and by providing users with interactive conversations. Despite the potential, LLMs face unique challenges in online shopping, such as domain-specific concepts, implicit knowledge, and heterogeneous user behaviors. Motivated by the potential and challenges, we propose Shopping MMLU, a diverse multi-task online shopping benchmark derived from real-world Amazon data. Shopping MMLU consists of 57 tasks covering 4 major shopping skills: concept understanding, knowledge reasoning, user behavior alignment, and multi-linguality, and can thus comprehensively evaluate the abilities of LLMs as general shop assistants. With Shopping MMLU, we benchmark over 20 existing LLMs and uncover valuable insights about practices and prospects of building versatile LLM-based shop assistants. Shopping MMLU can be publicly accessed at https://github.com/KL4805/ShoppingMMLU. In addition, with Shopping MMLU, we host a competition in KDD Cup 2024 with over 500 participating teams. The winning solutions and the associated workshop can be accessed at our website https://amazon-kddcup24.github.io/.

Evolutionary Contrastive Distillation for Language Model Alignment

The ability of large language models (LLMs) to execute complex instructions is essential for their real-world applications. However, several recent studies indicate that LLMs struggle with challenging instructions. In this paper, we propose Evolutionary Contrastive Distillation (ECD), a novel method for generating high-quality synthetic preference data designed to enhance the complex instruction-following capability of language models. ECD generates data that specifically illustrates the difference between a response that successfully follows a set of complex instructions and a response that is high-quality, but nevertheless makes some subtle mistakes. This is done by prompting LLMs to progressively evolve simple instructions to more complex instructions. When the complexity of an instruction is increased, the original successful response to the original instruction becomes a "hard negative" response for the new instruction, mostly meeting requirements of the new instruction, but barely missing one or two. By pairing a good response with such a hard negative response, and employing contrastive learning algorithms such as DPO, we improve language models' ability to follow complex instructions. Empirically, we observe that our method yields a 7B model that exceeds the complex instruction-following performance of current SOTA 7B models and is competitive even with open-source 70B models.

Uncertainty-Guided Enhancement on Driving Perception System via Foundation Models

Multimodal foundation models offer promising advancements for enhancing driving perception systems, but their high computational and financial costs pose challenges. We develop a method that leverages foundation models to refine predictions from existing driving perception models -- such as enhancing object classification accuracy -- while minimizing the frequency of using these resource-intensive models. The method quantitatively characterizes uncertainties in the perception model's predictions and engages the foundation model only when these uncertainties exceed a pre-specified threshold. Specifically, it characterizes uncertainty by calibrating the perception model's confidence scores into theoretical lower bounds on the probability of correct predictions using conformal prediction. Then, it sends images to the foundation model and queries for refining the predictions only if the theoretical bound of the perception model's outcome is below the threshold. Additionally, we propose a temporal inference mechanism that enhances prediction accuracy by integrating historical predictions, leading to tighter theoretical bounds. The method demonstrates a 10 to 15 percent improvement in prediction accuracy and reduces the number of queries to the foundation model by 50 percent, based on quantitative evaluations from driving datasets.

Enhancing Sequential Recommendations through Multi-Perspective Reflections and Iteration

Sequence recommendation (SeqRec) aims to predict the next item a user will interact with by understanding user intentions and leveraging collaborative filtering information. Large language models (LLMs) have shown great promise in recommendation tasks through prompt-based, fixed reflection libraries, and fine-tuning techniques. However, these methods face challenges, including lack of supervision, inability to optimize reflection sources, inflexibility to diverse user needs, and high computational costs. Despite promising results, current studies primarily focus on reflections of users' explicit preferences (e.g., item titles) while neglecting implicit preferences (e.g., brands) and collaborative filtering information. This oversight hinders the capture of preference shifts and dynamic user behaviors. Additionally, existing approaches lack mechanisms for reflection evaluation and iteration, often leading to suboptimal recommendations. To address these issues, we propose the Mixture of REflectors (MoRE) framework, designed to model and learn dynamic user preferences in SeqRec. Specifically, MoRE introduces three reflectors for generating LLM-based reflections on explicit preferences, implicit preferences, and collaborative signals. Each reflector incorporates a self-improving strategy, termed refining-and-iteration, to evaluate and iteratively update reflections. Furthermore, a meta-reflector employs a contextual bandit algorithm to select the most suitable expert and corresponding reflections for each user's recommendation, effectively capturing dynamic preferences. Extensive experiments on three real-world datasets demonstrate that MoRE consistently outperforms state-of-the-art methods, requiring less training time and GPU memory compared to other LLM-based approaches in SeqRec.

Unlock the Power of Frozen LLMs in Knowledge Graph Completion

Classical knowledge graph completion (KGC) methods rely solely on structural information, struggling with the inherent sparsity of knowledge graphs (KGs). Large Language Models (LLMs) learn extensive knowledge from large corpora with powerful context modeling, which is ideal for mitigating the limitations of previous methods. Directly fine-tuning LLMs offers great capability but comes at the cost of huge time and memory consumption, while utilizing frozen LLMs yields suboptimal results. In this work, we aim to leverage LLMs for KGC effectively and efficiently. We capture the context-aware hidden states of knowledge triples by employing prompts to stimulate the intermediate layers of LLMs. We then train a data-efficient classifier on these hidden states to harness the inherent capabilities of frozen LLMs in KGC. We also generate entity descriptions with subgraph sampling on KGs, reducing the ambiguity of triplets and enriching the knowledge representation. Extensive experiments on standard benchmarks showcase the efficiency and effectiveness of our approach. We outperform classical KGC methods on most datasets and match the performance of fine-tuned LLMs. Additionally, compared to fine-tuned LLMs, we boost GPU memory efficiency by \textbf{$188\times$} and speed up training+inference by \textbf{$13.48\times$}.

PanoFree: Tuning-Free Holistic Multi-view Image Generation with Cross-view Self-Guidance

Immersive scene generation, notably panorama creation, benefits significantly from the adaptation of large pre-trained text-to-image (T2I) models for multi-view image generation. Due to the high cost of acquiring multi-view images, tuning-free generation is preferred. However, existing methods are either limited to simple correspondences or require extensive fine-tuning to capture complex ones. We present PanoFree, a novel method for tuning-free multi-view image generation that supports an extensive array of correspondences. PanoFree sequentially generates multi-view images using iterative warping and inpainting, addressing the key issues of inconsistency and artifacts from error accumulation without the need for fine-tuning. It improves error accumulation by enhancing cross-view awareness and refines the warping and inpainting processes via cross-view guidance, risky area estimation and erasing, and symmetric bidirectional guided generation for loop closure, alongside guidance-based semantic and density control for scene structure preservation. In experiments on Planar, 360{\deg}, and Full Spherical Panoramas, PanoFree demonstrates significant error reduction, improves global consistency, and boosts image quality without extra fine-tuning. Compared to existing methods, PanoFree is up to 5x more efficient in time and 3x more efficient in GPU memory usage, and maintains superior diversity of results (2x better in our user study). PanoFree offers a viable alternative to costly fine-tuning or the use of additional pre-trained models. Project website at https://panofree.github.io/.

Robust Fast Adaptation from Adversarially Explicit Task Distribution Generation

Meta-learning is a practical learning paradigm to transfer skills across tasks from a few examples. Nevertheless, the existence of task distribution shifts tends to weaken meta-learners' generalization capability, particularly when the task distribution is naively hand-crafted or based on simple priors that fail to cover typical scenarios sufficiently. Here, we consider explicitly generative modeling task distributions placed over task identifiers and propose robustifying fast adaptation from adversarial training. Our approach, which can be interpreted as a model of a Stackelberg game, not only uncovers the task structure during problem-solving from an explicit generative model but also theoretically increases the adaptation robustness in worst cases. This work has practical implications, particularly in dealing with task distribution shifts in meta-learning, and contributes to theoretical insights in the field. Our method demonstrates its robustness in the presence of task subpopulation shifts and improved performance over SOTA baselines in extensive experiments. The project is available at https://sites.google.com/view/ar-metalearn.

Diffusion Models for Multi-Task Generative Modeling

Diffusion-based generative modeling has been achieving state-of-the-art results on various generation tasks. Most diffusion models, however, are limited to a single-generation modeling. Can we generalize diffusion models with the ability of multi-modal generative training for more generalizable modeling? In this paper, we propose a principled way to define a diffusion model by constructing a unified multi-modal diffusion model in a common diffusion space. We define the forward diffusion process to be driven by an information aggregation from multiple types of task-data, e.g., images for a generation task and labels for a classification task. In the reverse process, we enforce information sharing by parameterizing a shared backbone denoising network with additional modality-specific decoder heads. Such a structure can simultaneously learn to generate different types of multi-modal data with a multi-task loss, which is derived from a new multi-modal variational lower bound that generalizes the standard diffusion model. We propose several multimodal generation settings to verify our framework, including image transition, masked-image training, joint image-label and joint image-representation generative modeling. Extensive experimental results on ImageNet indicate the effectiveness of our framework for various multi-modal generative modeling, which we believe is an important research direction worthy of more future explorations.