We formalize the problem of prompt compression for large language models (LLMs) and present a framework to unify token-level prompt compression methods which create hard prompts for black-box models. We derive the distortion-rate function for this setup as a linear program, and provide an efficient algorithm to compute this fundamental limit via the dual of the linear program. Using the distortion-rate function as the baseline, we study the performance of existing compression schemes on a synthetic dataset consisting of prompts generated from a Markov chain, natural language queries, and their respective answers. Our empirical analysis demonstrates the criticality of query-aware prompt compression, where the compressor has knowledge of the downstream task/query for the black-box LLM. We show that there is a large gap between the performance of current prompt compression methods and the optimal strategy, and propose a query-aware, variable-rate adaptation of a prior work to close the gap. We extend our experiments to a small natural language dataset to further confirm our findings on our synthetic dataset.

As scaling laws in generative AI push performance, they also simultaneously concentrate the development of these models among actors with large computational resources. With a focus on text-to-image (T2I) generative models, we aim to address this bottleneck by demonstrating very low-cost training of large-scale T2I diffusion transformer models. As the computational cost of transformers increases with the number of patches in each image, we propose to randomly mask up to 75% of the image patches during training. We propose a deferred masking strategy that preprocesses all patches using a patch-mixer before masking, thus significantly reducing the performance degradation with masking, making it superior to model downscaling in reducing computational cost. We also incorporate the latest improvements in transformer architecture, such as the use of mixture-of-experts layers, to improve performance and further identify the critical benefit of using synthetic images in micro-budget training. Finally, using only 37M publicly available real and synthetic images, we train a 1.16 billion parameter sparse transformer with only \$1,890 economical cost and achieve a 12.7 FID in zero-shot generation on the COCO dataset. Notably, our model achieves competitive FID and high-quality generations while incurring 118$\times$ lower cost than stable diffusion models and 14$\times$ lower cost than the current state-of-the-art approach that costs \$28,400. We aim to release our end-to-end training pipeline to further democratize the training of large-scale diffusion models on micro-budgets.

Over summer 2024, the world will be looking at Paris to encourage their favorite athletes win the Olympic gold medal. In handball, few nations will fight hard to win the precious metal with speculations predicting the victory for France or Denmark for men and France or Norway for women. However, there is so far no scientific method proposed to predict the final results of the competition. In this work, we leverage a deep learning model to predict the results of the handball tournament of the 2024 Olympic Games. This model, coupled with explainable AI (xAI) techniques, allows us to extract insightful information about the main factors influencing the outcome of each match. Notably, xAI helps sports experts understand how factors like match information or individual athlete performance contribute to the predictions. Furthermore, we integrate Large Language Models (LLMs) to generate human-friendly explanations that highlight the most important factors impacting the match results. By providing human-centric explanations, our approach offers a deeper understanding of the AI predictions, making them more actionable for coaches and analysts.

The Industrial Control System (ICS) environment encompasses a wide range of intricate communication protocols, posing substantial challenges for Security Operations Center (SOC) analysts tasked with monitoring, interpreting, and addressing network activities and security incidents. Conventional monitoring tools and techniques often struggle to provide a clear understanding of the nature and intent of ICS-specific communications. To enhance comprehension, we propose a software solution powered by a Large Language Model (LLM). This solution currently focused on BACnet protocol, processes a packet file data and extracts context by using a mapping database, and contemporary context retrieval methods for Retrieval Augmented Generation (RAG). The processed packet information, combined with the extracted context, serves as input to the LLM, which generates a concise packet file summary for the user. The software delivers a clear, coherent, and easily understandable summary of network activities, enabling SOC analysts to better assess the current state of the control system.

Large multimodal models (LMMs) are processing increasingly longer and richer inputs. Albeit the progress, few public benchmark is available to measure such development. To mitigate this gap, we introduce LongVideoBench, a question-answering benchmark that features video-language interleaved inputs up to an hour long. Our benchmark includes 3,763 varying-length web-collected videos with their subtitles across diverse themes, designed to comprehensively evaluate LMMs on long-term multimodal understanding. To achieve this, we interpret the primary challenge as to accurately retrieve and reason over detailed multimodal information from long inputs. As such, we formulate a novel video question-answering task termed referring reasoning. Specifically, as part of the question, it contains a referring query that references related video contexts, called referred context. The model is then required to reason over relevant video details from the referred context. Following the paradigm of referring reasoning, we curate 6,678 human-annotated multiple-choice questions in 17 fine-grained categories, establishing one of the most comprehensive benchmarks for long-form video understanding. Evaluations suggest that the LongVideoBench presents significant challenges even for the most advanced proprietary models (e.g. GPT-4o, Gemini-1.5-Pro, GPT-4-Turbo), while their open-source counterparts show an even larger performance gap. In addition, our results indicate that model performance on the benchmark improves only when they are capable of processing more frames, positioning LongVideoBench as a valuable benchmark for evaluating future-generation long-context LMMs.

With the development of large language models (LLMs) like ChatGPT, both their vast applications and potential vulnerabilities have come to the forefront. While developers have integrated multiple safety mechanisms to mitigate their misuse, a risk remains, particularly when models encounter adversarial inputs. This study unveils an attack mechanism that capitalizes on human conversation strategies to extract harmful information from LLMs. We delineate three pivotal strategies: (i) decomposing malicious questions into seemingly innocent sub-questions; (ii) rewriting overtly malicious questions into more covert, benign-sounding ones; (iii) enhancing the harmfulness of responses by prompting models for illustrative examples. Unlike conventional methods that target explicit malicious responses, our approach delves deeper into the nature of the information provided in responses. Through our experiments conducted on GPT-3.5-turbo, GPT-4, and Llama2, our method has demonstrated a marked efficacy compared to conventional attack methods. In summary, this work introduces a novel attack method that outperforms previous approaches, raising an important question: How to discern whether the ultimate intent in a dialogue is malicious?

Large Language Models (LLMs) are widely used across various domains, processing millions of daily requests. This surge in demand poses significant challenges in optimizing throughput and latency while keeping costs manageable. The Key-Value (KV) cache, a standard method for retaining previous computations, makes LLM inference highly bounded by memory. While batching strategies can enhance performance, they frequently lead to significant memory fragmentation. Even though cutting-edge systems like vLLM mitigate KV cache fragmentation using paged Attention mechanisms, they still suffer from inefficient memory and computational operations due to the tightly coupled page management and computation kernels. This study introduces the vTensor, an innovative tensor structure for LLM inference based on GPU virtual memory management (VMM). vTensor addresses existing limitations by decoupling computation from memory defragmentation and offering dynamic extensibility. Our framework employs a CPU-GPU heterogeneous approach, ensuring efficient, fragmentation-free memory management while accommodating various computation kernels across different LLM architectures. Experimental results indicate that vTensor achieves an average speedup of 1.86x across different models, with up to 2.42x in multi-turn chat scenarios. Additionally, vTensor provides average speedups of 2.12x and 3.15x in kernel evaluation, reaching up to 3.92x and 3.27x compared to SGLang Triton prefix-prefilling kernels and vLLM paged Attention kernel, respectively. Furthermore, it frees approximately 71.25% (57GB) of memory on the NVIDIA A100 GPU compared to vLLM, enabling more memory-intensive workloads.

Language agents, built on top of language models (LMs), are systems that can interact with complex environments, such as the open web. In this work, we examine whether such agents can perform realistic and time-consuming tasks on the web, e.g., monitoring real-estate markets or locating relevant nearby businesses. We introduce AssistantBench, a challenging new benchmark consisting of 214 realistic tasks that can be automatically evaluated, covering different scenarios and domains. We find that AssistantBench exposes the limitations of current systems, including language models and retrieval-augmented language models, as no model reaches an accuracy of more than 25 points. While closed-book LMs perform well, they exhibit low precision since they tend to hallucinate facts. State-of-the-art web agents reach a score of near zero. Additionally, we introduce SeePlanAct (SPA), a new web agent that significantly outperforms previous agents, and an ensemble of SPA and closed-book models reaches the best overall performance. Moreover, we analyze failures of current systems and highlight that web navigation remains a major challenge.

Recent studies have increasingly applied natural language processing (NLP) to automatically extract experimental research data from the extensive battery materials literature. Despite the complex process involved in battery manufacturing -- from material synthesis to cell assembly -- there has been no comprehensive study systematically organizing this information. In response, we propose a language modeling-based protocol, Text-to-Battery Recipe (T2BR), for the automatic extraction of end-to-end battery recipes, validated using a case study on batteries containing LiFePO4 cathode material. We report machine learning-based paper filtering models, screening 2,174 relevant papers from the keyword-based search results, and unsupervised topic models to identify 2,876 paragraphs related to cathode synthesis and 2,958 paragraphs related to cell assembly. Then, focusing on the two topics, two deep learning-based named entity recognition models are developed to extract a total of 30 entities -- including precursors, active materials, and synthesis methods -- achieving F1 scores of 88.18% and 94.61%. The accurate extraction of entities enables the systematic generation of 165 end-toend recipes of LiFePO4 batteries. Our protocol and results offer valuable insights into specific trends, such as associations between precursor materials and synthesis methods, or combinations between different precursor materials. We anticipate that our findings will serve as a foundational knowledge base for facilitating battery-recipe information retrieval. The proposed protocol will significantly accelerate the review of battery material literature and catalyze innovations in battery design and development.

Web authors frequently embed social media to support and enrich their content, creating the potential to derive web-based, cross-platform social media representations that can enable more effective social media retrieval systems and richer scientific analyses. As step toward such capabilities, we introduce a novel language modeling framework that enables automatic annotation of roles that social media entities play in their embedded web context. Using related communication theory, we liken social media embeddings to quotes, formalize the page context as structured natural language signals, and identify a taxonomy of roles for quotes within the page context. We release SocialQuotes, a new data set built from the Common Crawl of over 32 million social quotes, 8.3k of them with crowdsourced quote annotations. Using SocialQuotes and the accompanying annotations, we provide a role classification case study, showing reasonable performance with modern-day LLMs, and exposing explainable aspects of our framework via page content ablations. We also classify a large batch of un-annotated quotes, revealing interesting cross-domain, cross-platform role distributions on the web.