The context caching technique is employed to accelerate the Multimodal Large Language Model (MLLM) inference by prevailing serving platforms currently. However, this approach merely reuses the Key-Value (KV) cache of the initial sequence of prompt, resulting in full KV cache recomputation even if the prefix differs slightly. This becomes particularly inefficient in the context of interleaved text and images, as well as multimodal retrieval-augmented generation. This paper proposes position-independent caching as a more effective approach for multimodal information management. We have designed and implemented a caching system, named MPIC, to address both system-level and algorithm-level challenges. MPIC stores the KV cache on local or remote disks when receiving multimodal data, and calculates and loads the KV cache in parallel during inference. To mitigate accuracy degradation, we have incorporated integrated reuse and recompute mechanisms within the system. The experimental results demonstrate that MPIC can achieve up to 54% reduction in response time compared to existing context caching systems, while maintaining negligible or no accuracy loss.
Recent advancements in Large Language Model (LLM) compression, such as BitNet and BitNet b1.58, have marked significant strides in reducing the computational demands of LLMs through innovative one-bit quantization techniques. We extend this frontier by looking at Large Inference Models (LIMs) that have become indispensable across various applications. However, their scale and complexity often come at a significant computational cost. We introduce a novel approach that leverages one-bit algorithm unrolling, effectively integrating information from the physical world in the model architecture. Our method achieves a bit-per-link rate significantly lower than the 1.58 bits reported in prior work, thanks to the natural sparsity that emerges in our network architectures. We numerically demonstrate that the proposed one-bit algorithm unrolling scheme can improve both training and test outcomes by effortlessly increasing the number of layers while substantially compressing the network. Additionally, we provide theoretical results on the generalization gap, convergence rate, stability, and sensitivity of our proposed one-bit algorithm unrolling.
Fine-tuning large language models (LLMs) is often constrained by the computational costs of processing massive datasets. We propose \textbf{QLESS} (Quantized Low-rank Gradient Similarity Search), which integrates gradient quantization with the LESS framework to enable memory-efficient data valuation and selection. QLESS employs a two-step compression process: first, it obtains low-dimensional gradient representations through LoRA-based random projection; then, it quantizes these gradients to low-bitwidth representations. Experiments on multiple LLM architectures (LLaMA, Mistral, Qwen) and benchmarks (MMLU, BBH, TyDiQA) show that QLESS achieves comparable data selection performance to LESS while reducing memory usage by up to 16x. Even 1-bit gradient quantization preserves data valuation quality. These findings underscore QLESS as a practical, scalable approach to identifying informative examples within strict memory constraints.
This paper presents a comparative study aimed at optimizing Llama2 inference, a critical aspect of machine learning and natural language processing (NLP). We evaluate various programming languages and frameworks, including TensorFlow, PyTorch, Python, Mojo, C++, and Java, analyzing their performance in terms of speed, memory consumption, and ease of implementation through extensive benchmarking. Strengths and limitations of each approach are highlighted, along with proposed optimization strategies for parallel processing and hardware utilization. Furthermore, we investigate the Mojo SDK, a novel framework designed for large language model (LLM) inference on Apple Silicon, benchmarking its performance against implementations in C, C++, Rust, Zig, Go, and Julia. Our experiments, conducted on an Apple M1 Max, demonstrate Mojo SDK's competitive performance, ease of use, and seamless Python compatibility, positioning it as a strong alternative for LLM inference on Apple Silicon. We also discuss broader implications for LLM deployment on resource-constrained hardware and identify potential directions for future research.
Large Language Models (LLM)-based systems, i.e. interconnected elements that include an LLM as a central component (e.g., conversational agents), are typically monolithic static architectures that rely on a single LLM for all user queries. However, they often require different preprocessing strategies, levels of reasoning, or knowledge. Generalist LLMs (e.g. GPT-4) trained on very large multi-topic corpora can perform well in a variety of tasks. They require significant financial, energy, and hardware resources that may not be justified for basic tasks. This implies potentially investing in unnecessary costs for a given query. To overcome this problem, a routing mechanism routes user queries to the most suitable components, such as smaller LLMs or experts in specific topics. This approach may improve response quality while minimising costs. Routing can be expanded to other components of the conversational agent architecture, such as the selection of optimal embedding strategies. This paper explores key considerations for integrating routing into LLM-based systems, focusing on resource management, cost definition, and strategy selection. Our main contributions include a formalisation of the problem, a novel taxonomy of existing approaches emphasising relevance and resource efficiency, and a comparative analysis of these strategies in relation to industry practices. Finally, we identify critical challenges and directions for future research.
Background: The global cost of drug discovery and development exceeds $200 billion annually, with clinical trial outcomes playing a critical role in the regulatory approval of new drugs and impacting patient outcomes. Despite their significance, large-scale, high-quality clinical trial outcome data are not readily available to the public, limiting advances in trial outcome predictive modeling. Methods: We introduce the Clinical Trial Outcome (CTO) knowledge base, a fully reproducible, large-scale (around 125K drug and biologics trials), open-source of clinical trial information including large language model (LLM) interpretations of publications, matched trials over phases, sentiment analysis from news, stock prices of trial sponsors, and other trial-related metrics. From this knowledge base, we additionally performed manual annotation of a set of recent clinical trials from 2020-2024. Results: We evaluated the quality of our knowledge base by generating high-quality trial outcome labels that demonstrate strong agreement with previously published expert annotations, achieving an F1 score of 94 for Phase 3 trials and 91 across all phases. Additionally, we benchmarked a suite of standard machine learning models on our manually annotated set, highlighting the distribution shift of recent trials and the need for continuously updated labeling methods. Conclusions: By analyzing CTO's performance on recent trials, we showed a need for recent, high-quality trial outcome labels. We release our knowledge base and labels to the public at https://chufangao.github.io/CTOD, which will also be regularly updated to support ongoing research in clinical trial outcomes, offering insights that could optimize the drug development process.
State of the art large language models rely on randomization to respond to a prompt. As an immediate consequence, a model may respond differently to the same prompt if asked multiple times. In this work, we argue that the evaluation and ranking of large language models should control for the randomization underpinning their functioning. Our starting point is the development of a causal model for coupled autoregressive generation, which allows different large language models to sample responses with the same source of randomness. Building upon our causal model, we first show that, on evaluations based on benchmark datasets, coupled autoregressive generation leads to the same conclusions as vanilla autoregressive generation but using provably fewer samples. However, we further show that, on evaluations based on (human) pairwise comparisons, coupled and vanilla autoregressive generation can surprisingly lead to different rankings when comparing more than two models, even with an infinite amount of samples. This suggests that the apparent advantage of a model over others in existing evaluation protocols may not be genuine but rather confounded by the randomness inherent to the generation process. To illustrate and complement our theoretical results, we conduct experiments with several large language models from the Llama family. We find that, across multiple knowledge areas from the popular MMLU benchmark dataset, coupled autoregressive generation requires up to 40% fewer samples to reach the same conclusions as vanilla autoregressive generation. Further, using data from the LMSYS Chatbot Arena platform, we find that the win-rates derived from pairwise comparisons by a strong large language model to prompts differ under coupled and vanilla autoregressive generation.
Reliable extraction of structured data from radiology reports using Large Language Models (LLMs) remains challenging, especially for complex, non-English texts like Hebrew. This study introduces an agent-based uncertainty-aware approach to improve the trustworthiness of LLM predictions in medical applications. We analyzed 9,683 Hebrew radiology reports from Crohn's disease patients (from 2010 to 2023) across three medical centers. A subset of 512 reports was manually annotated for six gastrointestinal organs and 15 pathological findings, while the remaining reports were automatically annotated using HSMP-BERT. Structured data extraction was performed using Llama 3.1 (Llama 3-8b-instruct) with Bayesian Prompt Ensembles (BayesPE), which employed six semantically equivalent prompts to estimate uncertainty. An Agent-Based Decision Model integrated multiple prompt outputs into five confidence levels for calibrated uncertainty and was compared against three entropy-based models. Performance was evaluated using accuracy, F1 score, precision, recall, and Cohen's Kappa before and after filtering high-uncertainty cases. The agent-based model outperformed the baseline across all metrics, achieving an F1 score of 0.3967, recall of 0.6437, and Cohen's Kappa of 0.3006. After filtering high-uncertainty cases (greater than or equal to 0.5), the F1 score improved to 0.4787, and Kappa increased to 0.4258. Uncertainty histograms demonstrated clear separation between correct and incorrect predictions, with the agent-based model providing the most well-calibrated uncertainty estimates. By incorporating uncertainty-aware prompt ensembles and an agent-based decision model, this approach enhances the performance and reliability of LLMs in structured data extraction from radiology reports, offering a more interpretable and trustworthy solution for high-stakes medical applications.
Ensemble methods enhance Large Language Models (LLMs) by combining multiple models but suffer from high computational costs. In this paper, we introduce Speculative Ensemble, a novel framework that accelerates LLM ensembles without sacrificing performance, inspired by Speculative Decoding-where a small proposal model generates tokens sequentially, and a larger target model verifies them in parallel. Our approach builds on two key insights: (1) the verification distribution can be the ensemble distribution of both the proposal and target models, and (2) alternating each model as the proposer and verifier can further enhance efficiency. We generalize this method to ensembles with n models and theoretically prove that SE is never slower than a standard ensemble, typically achieving faster speed. Extensive experiments demonstrate speed improvements of 1.11x-2.23x over standard ensemble techniques without compromising generation quality. Our code is available at https://github.com/Kamichanw/Speculative-Ensemble/
Real-world applications are stretching context windows to hundreds of thousand of tokens while Large Language Models (LLMs) swell from billions to trillions of parameters. This dual expansion send compute and memory costs skyrocketing, making token compression indispensable. We introduce Vision Centric Token Compression (Vist), a slow-fast compression framework that mirrors human reading: the fast path renders distant tokens into images, letting a frozen, lightweight vision encoder skim the low-salience context; the slow path feeds the proximal window into the LLM for fine-grained reasoning. A Probability-Informed Visual Enhancement (PVE) objective masks high-frequency tokens during training, steering the Resampler to concentrate on semantically rich regions-just as skilled reader gloss over function words. On eleven in-context learning benchmarks, Vist achieves the same accuracy with 2.3 times fewer tokens, cutting FLOPs by 16% and memory by 50%. This method delivers remarkable results, outperforming the strongest text encoder-based compression method CEPE by 7.6% on average over benchmarks like TriviaQA, NQ, PopQA, NLUI, and CLIN, setting a new standard for token efficiency in LLMs. The source code will be released.
Video foundation models have achieved significant advancement with the help of network upgrade as well as model scale-up. However, they are still hard to meet requirements of applications due to unsatisfied generation quality. To solve this problem, we propose to align video foundation models with human preferences from the perspective of post-training in this paper. Consequently, we introduce an Iterative Preference Optimization strategy to enhance generated video quality by incorporating human feedback. Specifically, IPO exploits a critic model to justify video generations for pairwise ranking as in Direct Preference Optimization or point-wise scoring as in Kahneman-Tversky Optimization. Given this, IPO optimizes video foundation models with guidance of signals from preference feedback, which helps improve generated video quality in subject consistency, motion smoothness and aesthetic quality, etc. In addition, IPO incorporates the critic model with the multi-modality large language model, which enables it to automatically assign preference labels without need of retraining or relabeling. In this way, IPO can efficiently perform multi-round preference optimization in an iterative manner, without the need of tediously manual labeling. Comprehensive experiments demonstrate that the proposed IPO can effectively improve the video generation quality of a pretrained model and help a model with only 2B parameters surpass the one with 5B parameters. Besides, IPO achieves new state-of-the-art performance on VBench benchmark. We will release our source codes, models as well as dataset to advance future research and applications.
Large vision-and-language models (LVLMs) typically treat visual and textual embeddings as homogeneous inputs to a large language model (LLM). However, these inputs are inherently different: visual inputs are multi-dimensional and contextually rich, often pre-encoded by models like CLIP, while textual inputs lack this structure. In this paper, we propose Decomposed Attention (D-Attn), a novel method that processes visual and textual embeddings differently by decomposing the 1-D causal self-attention in LVLMs. After the attention decomposition, D-Attn diagonalizes visual-to-visual self-attention, reducing computation from $\mathcal{O}(|V|^2)$ to $\mathcal{O}(|V|)$ for $|V|$ visual embeddings without compromising performance. Moreover, D-Attn debiases positional encodings in textual-to-visual cross-attention, further enhancing visual understanding. Finally, we introduce an $\alpha$-weighting strategy to merge visual and textual information, maximally preserving the pre-trained LLM's capabilities with minimal modifications. Extensive experiments and rigorous analyses validate the effectiveness of D-Attn, demonstrating significant improvements on multiple image benchmarks while significantly reducing computational costs. Code, data, and models will be publicly available.
This paper addresses the critical disconnect between prediction and decision quality in portfolio optimization by integrating Large Language Models (LLMs) with decision-focused learning. We demonstrate both theoretically and empirically that minimizing the prediction error alone leads to suboptimal portfolio decisions. We aim to exploit the representational power of LLMs for investment decisions. An attention mechanism processes asset relationships, temporal dependencies, and macro variables, which are then directly integrated into a portfolio optimization layer. This enables the model to capture complex market dynamics and align predictions with the decision objectives. Extensive experiments on S\&P100 and DOW30 datasets show that our model consistently outperforms state-of-the-art deep learning models. In addition, gradient-based analyses show that our model prioritizes the assets most crucial to decision making, thus mitigating the effects of prediction errors on portfolio performance. These findings underscore the value of integrating decision objectives into predictions for more robust and context-aware portfolio management.
Deep learning (DL) has made notable progress in addressing complex radio access network control challenges that conventional analytic methods have struggled to solve. However, DL has shown limitations in solving constrained NP-hard problems often encountered in network optimization, such as those involving quality of service (QoS) or discrete variables like user indices. Current solutions rely on domain-specific architectures or heuristic techniques, and a general DL approach for constrained optimization remains undeveloped. Moreover, even minor changes in communication objectives demand time-consuming retraining, limiting their adaptability to dynamic environments where task objectives, constraints, environmental factors, and communication scenarios frequently change. To address these challenges, we propose a large language model for resource allocation optimizer (LLM-RAO), a novel approach that harnesses the capabilities of LLMs to address the complex resource allocation problem while adhering to QoS constraints. By employing a prompt-based tuning strategy to flexibly convey ever-changing task descriptions and requirements to the LLM, LLM-RAO demonstrates robust performance and seamless adaptability in dynamic environments without requiring extensive retraining. Simulation results reveal that LLM-RAO achieves up to a 40% performance enhancement compared to conventional DL methods and up to an $80$\% improvement over analytical approaches. Moreover, in scenarios with fluctuating communication objectives, LLM-RAO attains up to 2.9 times the performance of traditional DL-based networks.
Bug reports often lack sufficient detail for developers to reproduce and fix the underlying defects. Bug Reproduction Tests (BRTs), tests that fail when the bug is present and pass when it has been resolved, are crucial for debugging, but they are rarely included in bug reports, both in open-source and in industrial settings. Thus, automatically generating BRTs from bug reports has the potential to accelerate the debugging process and lower time to repair. This paper investigates automated BRT generation within an industry setting, specifically at Google, focusing on the challenges of a large-scale, proprietary codebase and considering real-world industry bugs extracted from Google's internal issue tracker. We adapt and evaluate a state-of-the-art BRT generation technique, LIBRO, and present our agent-based approach, BRT Agent, which makes use of a fine-tuned Large Language Model (LLM) for code editing. Our BRT Agent significantly outperforms LIBRO, achieving a 28% plausible BRT generation rate, compared to 10% by LIBRO, on 80 human-reported bugs from Google's internal issue tracker. We further investigate the practical value of generated BRTs by integrating them with an Automated Program Repair (APR) system at Google. Our results show that providing BRTs to the APR system results in 30% more bugs with plausible fixes. Additionally, we introduce Ensemble Pass Rate (EPR), a metric which leverages the generated BRTs to select the most promising fixes from all fixes generated by APR system. Our evaluation on EPR for Top-K and threshold-based fix selections demonstrates promising results and trade-offs. For example, EPR correctly selects a plausible fix from a pool of 20 candidates in 70% of cases, based on its top-1 ranking.
Cass Sunstein's essay 'On Bob Dylan' describes Dylan's 'dishabituating' style -- a constant refusal to conform to expectation and a penchant for reinventing his musical and lyrical identity. In this paper, I extend Sunstein's observations through a large-scale computational analysis of Dylan's lyrics from 1962 to 2012. Using o3-mini-high (a large language model), I extract concept-to-concept relationships from the lyrics and construct directed knowledge graphs that capture Dylan's thematic structure. I then quantify shifts in sentiment, metaphorical expression, thematic diversity, and network complexity over time. The results indicate that Dylan's lyrics increasingly rely on metaphor, display an evolving sentiment profile, and exhibit heightened dishabituation -- measured here as a growing variance in the network centrality of key concepts. I also find that references to movement, protest, and mythic imagery fluctuate in ways that align with well-known phases of Dylan's career, reflecting the dynamic and unpredictable quality of his art. These findings not only deepen our empirical understanding of Sunstein's thesis but also introduce a novel computational method for analyzing an artist's evolution-offering broader applicability to the study of cultural and creative change.
Unit tests (UTs) play an instrumental role in assessing code correctness as well as providing feedback to large language models (LLMs), motivating automated test generation. However, we uncover a trade-off between generating unit test inputs that reveal errors when given a faulty code and correctly predicting the unit test output without access to the gold solution. To address this trade-off, we propose UTGen, which teaches LLMs to generate unit test inputs that reveal errors along with their correct expected outputs based on task descriptions. Since model-generated tests can provide noisy signals (e.g., from incorrectly predicted outputs), we propose UTDebug that (i) scales UTGen via test-time compute to improve UT output prediction, and (ii) validates and backtracks edits based on multiple generated UTs to avoid overfitting, and helps LLMs debug effectively. We show that UTGen outperforms other LLM-based baselines by 7.59% based on a metric measuring the presence of both error-revealing UT inputs and correct UT outputs. When used with UTDebug, we find that feedback from UTGen's unit tests improves pass@1 accuracy of Qwen2.5 32B on HumanEvalFix and our own harder debugging split of MBPP+ by over 3.17% and 12.35% (respectively) over other LLM-based UT generation baselines. Lastly, we demonstrate that UTGen is a better judge for code correctness, outperforming a state-of-the-art trained 8B reward model by 4.43% on HumanEval+ with best-of-10 sampling using Qwen2.5 7B.
With the rise of social media, misinformation has become increasingly prevalent, fueled largely by the spread of rumors. This study explores the use of Large Language Model (LLM) agents within a novel framework to simulate and analyze the dynamics of rumor propagation across social networks. To this end, we design a variety of LLM-based agent types and construct four distinct network structures to conduct these simulations. Our framework assesses the effectiveness of different network constructions and agent behaviors in influencing the spread of rumors. Our results demonstrate that the framework can simulate rumor spreading across more than one hundred agents in various networks with thousands of edges. The evaluations indicate that network structure, personas, and spreading schemes can significantly influence rumor dissemination, ranging from no spread to affecting 83\% of agents in iterations, thereby offering a realistic simulation of rumor spread in social networks.
Computational Fluid Dynamics (CFD) is widely used in aerospace, energy, and biology to model fluid flow, heat transfer, and chemical reactions. While Large Language Models (LLMs) have transformed various domains, their application in CFD remains limited, particularly for complex tasks like post-processing. To bridge this gap, we introduce MetaOpenFOAM 2.0, which leverages Chain of Thought (COT) decomposition and iterative verification to enhance accessibility for non-expert users through natural language inputs. Tested on a new benchmark covering simulation (fluid flow, heat transfer, combustion) and post-processing (extraction, visualization), MetaOpenFOAM 2.0 achieved an Executability score of 6.3/7 and a pass rate of 86.9%, significantly outperforming MetaOpenFOAM 1.0 (2.1/7, 0%). Additionally, it proved cost-efficient, averaging $0.15 per case. An ablation study confirmed that COT-driven decomposition and iterative refinement substantially improved task performance. Furthermore, scaling laws showed that increasing COT steps enhanced accuracy while raising token usage, aligning with LLM post-training scaling trends. These results highlight the transformative potential of LLMs in automating CFD workflows for industrial and research applications. Code is available at https://github.com/Terry-cyx/MetaOpenFOAM
Recent developments in Large Language Model (LLM) agents are revolutionizing Autonomous Software Engineering (ASE), enabling automated coding, problem fixes, and feature improvements. However, localization -- precisely identifying software problems by navigating to relevant code sections -- remains a significant challenge. Current approaches often yield suboptimal results due to a lack of effective integration between LLM agents and precise code search mechanisms. This paper introduces OrcaLoca, an LLM agent framework that improves accuracy for software issue localization by integrating priority-based scheduling for LLM-guided action, action decomposition with relevance scoring, and distance-aware context pruning. Experimental results demonstrate that OrcaLoca becomes the new open-source state-of-the-art (SOTA) in function match rate (65.33%) on SWE-bench Lite. It also improves the final resolved rate of an open-source framework by 6.33 percentage points through its patch generation integration.