Misgendering is the act of referring to someone by a gender that does not match their chosen identity. It marginalizes and undermines a person's sense of self, causing significant harm. English-based approaches have clear-cut approaches to avoiding misgendering, such as the use of the pronoun ``they''. However, other languages pose unique challenges due to both grammatical and cultural constructs. In this work we develop methodologies to assess and mitigate misgendering across 42 languages and dialects using a participatory-design approach to design effective and appropriate guardrails across all languages. We test these guardrails in a standard LLM-based application (meeting transcript summarization), where both the data generation and the annotation steps followed a human-in-the-loop approach. We find that the proposed guardrails are very effective in reducing misgendering rates across all languages in the summaries generated, and without incurring loss of quality. Our human-in-the-loop approach demonstrates a method to feasibly scale inclusive and responsible AI-based solutions across multiple languages and cultures. We release the guardrails and synthetic dataset encompassing 42 languages, along with human and LLM-judge evaluations, to encourage further research on this subject.
Distillation has shown remarkable success in transferring knowledge from a Large Language Model (LLM) teacher to a student LLM. However, current distillation methods predominantly require the same tokenizer between the teacher and the student, restricting their applicability to only a small subset of teacher-student pairs. In this work, we develop a cross-tokenizer distillation method to solve this crucial deficiency. Our method is the first to enable cross-tokenizer distillation without a next-token prediction loss as the main objective, instead purely maximizing the student predictions' similarity to the teacher's predictions (known as pure distillation), while also being robust to large mismatches between the teacher and the student tokenizer function and vocabulary. Empirically, our method enables substantially improved performance as tested on two use cases. First, we show that viewing tokenizer transfer as self-distillation enables unprecedently effective transfer across tokenizers. We transfer (subword-level) Llama and Gemma models to byte-level tokenization more effectively than prior methods transfer to a similar subword tokenizer under a comparable training budget. Transferring different base models to the same tokenizer also enables ensembling them (e.g., via averaging their predicted probabilities) which boosts performance. Second, we use our cross-tokenizer distillation method to distil a large maths-specialized LLM into a smaller model, achieving competitive maths problem-solving performance. Overall, our results make substantial strides toward better adaptability and enhanced interaction between different LLMs.
In large language model (LLM) serving systems, executing each request consists of two phases: the compute-intensive prefill phase and the memory-intensive decoding phase. To prevent performance interference between the two phases, current LLM serving systems typically adopt prefill-decoding disaggregation, where the two phases are split across separate machines. However, we observe this approach leads to significant resource underutilization. Specifically, prefill instances that are compute-intensive suffer from low memory utilization, while decoding instances that are memory-intensive experience low compute utilization. To address this problem, this paper proposes Adrenaline, an attention disaggregation and offloading mechanism designed to enhance resource utilization and performance in LLM serving systems. Adrenaline's key innovation lies in disaggregating part of the attention computation in the decoding phase and offloading them to prefill instances. The memory-bound nature of decoding-phase attention computation inherently enables an effective offloading strategy, yielding two complementary advantages: 1) improved memory capacity and bandwidth utilization in prefill instances, and 2) increased decoding batch sizes that enhance compute utilization in decoding instances, collectively boosting overall system performance. Adrenaline achieves these gains through three key techniques: low-latency decoding synchronization, resource-efficient prefill colocation, and load-aware offloading scheduling. Experimental results show that Adrenaline achieves 2.28x higher memory capacity and 2.07x better memory bandwidth utilization in prefill instances, up to 1.67x improvements in compute utilization for decoding instances, and 1.68x higher overall inference throughput compared to state-of-the-art systems.
Composed Image Retrieval (CIR) is a complex task that aims to retrieve images based on a multimodal query. Typical training data consists of triplets containing a reference image, a textual description of desired modifications, and the target image, which are expensive and time-consuming to acquire. The scarcity of CIR datasets has led to zero-shot approaches utilizing synthetic triplets or leveraging vision-language models (VLMs) with ubiquitous web-crawled image-caption pairs. However, these methods have significant limitations: synthetic triplets suffer from limited scale, lack of diversity, and unnatural modification text, while image-caption pairs hinder joint embedding learning of the multimodal query due to the absence of triplet data. Moreover, existing approaches struggle with complex and nuanced modification texts that demand sophisticated fusion and understanding of vision and language modalities. We present CoLLM, a one-stop framework that effectively addresses these limitations. Our approach generates triplets on-the-fly from image-caption pairs, enabling supervised training without manual annotation. We leverage Large Language Models (LLMs) to generate joint embeddings of reference images and modification texts, facilitating deeper multimodal fusion. Additionally, we introduce Multi-Text CIR (MTCIR), a large-scale dataset comprising 3.4M samples, and refine existing CIR benchmarks (CIRR and Fashion-IQ) to enhance evaluation reliability. Experimental results demonstrate that CoLLM achieves state-of-the-art performance across multiple CIR benchmarks and settings. MTCIR yields competitive results, with up to 15% performance improvement. Our refined benchmarks provide more reliable evaluation metrics for CIR models, contributing to the advancement of this important field.
As creative writing tasks do not have singular correct answers, large language models (LLMs) trained to perform these tasks should be able to generate diverse valid outputs. However, LLM post-training often focuses on improving generation quality but neglects to facilitate output diversity. Hence, in creative writing generation, we investigate post-training approaches to promote both output diversity and quality. Our core idea is to include deviation -- the degree of difference between a training sample and all other samples with the same prompt -- in the training objective to facilitate learning from rare high-quality instances. By adopting our approach to direct preference optimization (DPO) and odds ratio preference optimization (ORPO), we demonstrate that we can promote the output diversity of trained models while minimally decreasing quality. Our best model with 8B parameters could achieve on-par diversity as a human-created dataset while having output quality similar to the best instruction-tuned models we examined, GPT-4o and DeepSeek-R1. We further validate our approaches with a human evaluation, an ablation, and a comparison to an existing diversification approach, DivPO.
Tibetan, a minority language in China, features a highly intricate grammatical structure, characterized by four verb tenses and a tense system with frequent irregularities, contributing to its extensive inflectional diversity. Recently, advances in Large Language Models (LLMs) have transformed the paradigm in many domains. Despite the success in other fields, current LLMs often fall short in catering to the needs of domain experts like Tibetans, and the potential of LLMs for Tibetan culture is under-explored. The intrinsic reasons are the immense and intricate nature of Tibetan culture as well as the necessity for higher granularity and richness in knowledge. Simultaneously, the complexity and uniqueness of its grammatical structure, coupled with its status as a minority ethnic language, contribute to data scarcity, which remains a fundamental challenge. To alleviate these issues, we introduce Llama-Sunshine (Sun-Shine), the first large language model for Tibetan culture, which is expert in various Tibetan language processing tasks. Sun-Shine incorporates state-of-the-art model architectures optimized for Tibetan's linguistic features. We also propose TIB-STC, a comprehensive dataset comprising diverse Tibetan texts such as literature, religious scripts, news, and conversational data, which is also the first large-scale dataset for Tibetan culture. Though comprehensive experiments, Sun-Shine not only demonstrates a higher level of knowledge expertise for Tibetan culture but also gains preliminary embodied intelligence capabilities in Tibetan language processing tasks, like language modeling, text classification, machine translation, and syntactic analysis. Moreover, it excels in low-resource scenarios, showcasing strong generalization capabilities.
The automatic generation of brain CT reports has gained widespread attention, given its potential to assist radiologists in diagnosing cranial diseases. However, brain CT scans involve extensive medical entities, such as diverse anatomy regions and lesions, exhibiting highly inconsistent spatial patterns in 3D volumetric space. This leads to biased learning of medical entities in existing methods, resulting in repetitiveness and inaccuracy in generated reports. To this end, we propose a Medical Entity-balanced Prompting Network (MEPNet), which harnesses the large language model (LLM) to fairly interpret various entities for accurate brain CT report generation. By introducing the visual embedding and the learning status of medical entities as enriched clues, our method prompts the LLM to balance the learning of diverse entities, thereby enhancing reports with comprehensive findings. First, to extract visual embedding of entities, we propose Knowledge-driven Joint Attention to explore and distill entity patterns using both explicit and implicit medical knowledge. Then, a Learning Status Scorer is designed to evaluate the learning of entity visual embeddings, resulting in unique learning status for individual entities. Finally, these entity visual embeddings and status are elaborately integrated into multi-modal prompts, to guide the text generation of LLM. This process allows LLM to self-adapt the learning process for biased-fitted entities, thereby covering detailed findings in generated reports. We conduct experiments on two brain CT report generation benchmarks, showing the effectiveness in clinical accuracy and text coherence.
Language models (LMs) are machine learning models designed to predict linguistic patterns by estimating the probability of word sequences based on large-scale datasets, such as text. LMs have a wide range of applications in natural language processing (NLP) tasks, including autocomplete and machine translation. Although larger datasets typically enhance LM performance, scalability remains a challenge due to constraints in computational power and resources. Distributed computing strategies offer essential solutions for improving scalability and managing the growing computational demand. Further, the use of sensitive datasets in training and deployment raises significant privacy concerns. Recent research has focused on developing decentralized techniques to enable distributed training and inference while utilizing diverse computational resources and enabling edge AI. This paper presents a survey on distributed solutions for various LMs, including large language models (LLMs), vision language models (VLMs), multimodal LLMs (MLLMs), and small language models (SLMs). While LLMs focus on processing and generating text, MLLMs are designed to handle multiple modalities of data (e.g., text, images, and audio) and to integrate them for broader applications. To this end, this paper reviews key advancements across the MLLM pipeline, including distributed training, inference, fine-tuning, and deployment, while also identifying the contributions, limitations, and future areas of improvement. Further, it categorizes the literature based on six primary focus areas of decentralization. Our analysis describes gaps in current methodologies for enabling distributed solutions for LMs and outline future research directions, emphasizing the need for novel solutions to enhance the robustness and applicability of distributed LMs.
The outcome of Large Language Model (LLM) pre-training strongly depends on weight initialization and variance control strategies. Although the importance of initial variance control has been well documented in neural networks in general, the literature on initialization and management of its growth during LLM pre-training, specifically, is somewhat sparse. In this paper, we introduce the Layer Index Rescaling (LIR) weight initialization scheme, and the Target Variance Rescaling (TVR) variance control strategy. Experiments on a 1B parameter LLaMA model demonstrate that better variance management using these techniques yields substantial improvements in downstream task performance (up to 4.6% on common pre-training benchmarks) and reduces extreme activation values, thus mitigating challenges associated with quantization and low-precision training. Our code is available at: https://github.com/bluorion-com/weight_rescaling.
Large language model (LLM) agents are increasingly capable of autonomously conducting cyberattacks, posing significant threats to existing applications. This growing risk highlights the urgent need for a real-world benchmark to evaluate the ability of LLM agents to exploit web application vulnerabilities. However, existing benchmarks fall short as they are limited to abstracted Capture the Flag competitions or lack comprehensive coverage. Building a benchmark for real-world vulnerabilities involves both specialized expertise to reproduce exploits and a systematic approach to evaluating unpredictable threats. To address this challenge, we introduce CVE-Bench, a real-world cybersecurity benchmark based on critical-severity Common Vulnerabilities and Exposures. In CVE-Bench, we design a sandbox framework that enables LLM agents to exploit vulnerable web applications in scenarios that mimic real-world conditions, while also providing effective evaluation of their exploits. Our evaluation shows that the state-of-the-art agent framework can resolve up to 13% of vulnerabilities.
Large language model (LLM) agents need to perform multi-turn interactions in real-world tasks. However, existing multi-turn RL algorithms for optimizing LLM agents fail to perform effective credit assignment over multiple turns while leveraging the generalization capabilities of LLMs and it remains unclear how to develop such algorithms. To study this, we first introduce a new benchmark, ColBench, where an LLM agent interacts with a human collaborator over multiple turns to solve realistic tasks in backend programming and frontend design. Building on this benchmark, we propose a novel RL algorithm, SWEET-RL (RL with Step-WisE Evaluation from Training-time information), that uses a carefully designed optimization objective to train a critic model with access to additional training-time information. The critic provides step-level rewards for improving the policy model. Our experiments demonstrate that SWEET-RL achieves a 6% absolute improvement in success and win rates on ColBench compared to other state-of-the-art multi-turn RL algorithms, enabling Llama-3.1-8B to match or exceed the performance of GPT4-o in realistic collaborative content creation.
Despite significant advancements, the practical deployment of Large Language Models (LLMs) is often hampered by their immense sizes, highlighting the need for effective compression techniques. Singular Value Decomposition (SVD) is a promising LLM compression technique. However, existing SVD-based compression methods fall short in reducing truncation losses, leading to less competitive performance in compressed models. In this work, we introduce SVD-LLM V2, a SVD-based LLM compression method that optimizes singular value truncation in SVD compression with two techniques. First, SVD-LLM V2 proposes to use theoretical truncation loss of weight matrices to assign a unique compression ratio to each weight matrix at different layers to accommodate weight redundancy heterogeneity. Second, SVD-LLM V2 proposes loss-optimized weight truncation to ensure that the truncated singular values result in a lower and more stable truncation loss in practice. We evaluate SVD-LLM V2 on ten datasets and five LLMs at various scales. Our results show SVD-LLM V2 outperforms state-of-the-art SVD-based LLM compression methods. Our code is available at https://github.com/AIoT-MLSys-Lab/SVD-LLM
Benchmark Data Contamination (BDC)-the inclusion of benchmark testing samples in the training set-has raised increasing concerns in Large Language Model (LLM) evaluation, leading to falsely inflated performance estimates and undermining evaluation reliability. To address this, researchers have proposed various mitigation strategies to update existing benchmarks, including modifying original questions or generating new ones based on them. However, a rigorous examination of the effectiveness of these mitigation strategies remains lacking. In this paper, we design a systematic and controlled pipeline along with two novel metrics-fidelity and contamination resistance-to provide a fine-grained and comprehensive assessment of existing BDC mitigation strategies. Previous assessment methods, such as accuracy drop and accuracy matching, focus solely on aggregate accuracy, often leading to incomplete or misleading conclusions. Our metrics address this limitation by emphasizing question-level evaluation result matching. Extensive experiments with 10 LLMs, 5 benchmarks, 20 BDC mitigation strategies, and 2 contamination scenarios reveal that no existing strategy significantly improves resistance over the vanilla case (i.e., no benchmark update) across all benchmarks, and none effectively balances fidelity and contamination resistance. These findings underscore the urgent need for designing more effective BDC mitigation strategies. Our code repository is available at https://github.com/ASTRAL-Group/BDC_mitigation_assessment.
Reasoning large language models are rapidly evolving across various domains. However, their capabilities in handling complex financial tasks still require in-depth exploration. In this paper, we introduce Fin-R1, a reasoning large language model specifically designed for the financial sector. Fin-R1 is built using a two-stage architecture, leveraging a financial reasoning dataset distilled and processed based on DeepSeek-R1. Through supervised fine-tuning (SFT) and reinforcement learning (RL) training, it demonstrates performance close to DeepSeek-R1 with a parameter size of 7 billion across a range of financial reasoning tasks. It achieves the state-of-the-art (SOTA) in the FinQA and ConvFinQA tasks between those LLMs in our evaluation, surpassing larger models in other tasks as well. Fin-R1 showcases strong reasoning and decision-making capabilities, providing solutions to various problems encountered in the financial domain. Our code is available at https://github.com/SUFE-AIFLM-Lab/Fin-R1.
The large language model (LLM)-as-judge paradigm has been used to meet the demand for a cheap, reliable, and fast evaluation of model outputs during AI system development and post-deployment monitoring. While judge models -- LLMs finetuned to specialize in assessing and critiquing model outputs -- have been touted as general purpose evaluators, they are typically evaluated only on non-contextual scenarios, such as instruction following. The omission of contextual settings -- those where external information is used as context to generate an output -- is surprising given the increasing prevalence of retrieval-augmented generation (RAG) and summarization use cases. Contextual assessment is uniquely challenging, as evaluation often depends on practitioner priorities, leading to conditional evaluation criteria (e.g., comparing responses based on factuality and then considering completeness if they are equally factual). To address the gap, we propose ContextualJudgeBench, a judge benchmark with 2,000 challenging response pairs across eight splits inspired by real-world contextual evaluation scenarios. We build our benchmark with a multi-pronged data construction pipeline that leverages both existing human annotations and model-based perturbations. Our comprehensive study across 11 judge models and 9 general purpose models, reveals that the contextual information and its assessment criteria present a significant challenge to even state-of-the-art models. For example, OpenAI's o1, the best-performing model, barely reaches 55% consistent accuracy.
With the rapid development of Large Language Models (LLMs), LLM-based agents have been widely adopted in various fields, becoming essential for autonomous decision-making and interactive tasks. However, current work typically relies on prompt design or fine-tuning strategies applied to vanilla LLMs, which often leads to limited effectiveness or suboptimal performance in complex agent-related environments. Although LLM optimization techniques can improve model performance across many general tasks, they lack specialized optimization towards critical agent functionalities such as long-term planning, dynamic environmental interaction, and complex decision-making. Although numerous recent studies have explored various strategies to optimize LLM-based agents for complex agent tasks, a systematic review summarizing and comparing these methods from a holistic perspective is still lacking. In this survey, we provide a comprehensive review of LLM-based agent optimization approaches, categorizing them into parameter-driven and parameter-free methods. We first focus on parameter-driven optimization, covering fine-tuning-based optimization, reinforcement learning-based optimization, and hybrid strategies, analyzing key aspects such as trajectory data construction, fine-tuning techniques, reward function design, and optimization algorithms. Additionally, we briefly discuss parameter-free strategies that optimize agent behavior through prompt engineering and external knowledge retrieval. Finally, we summarize the datasets and benchmarks used for evaluation and tuning, review key applications of LLM-based agents, and discuss major challenges and promising future directions. Our repository for related references is available at https://github.com/YoungDubbyDu/LLM-Agent-Optimization.
Traffic Signal Control (TSC) plays a critical role in urban traffic management by optimizing traffic flow and mitigating congestion. While Large Language Models (LLMs) have recently emerged as promising tools for TSC due to their exceptional problem-solving and generalization capabilities, existing approaches fail to address the essential need for inter-agent coordination, limiting their effectiveness in achieving network-wide optimization. To bridge this gap, we propose CoLLMLight, a cooperative LLM agent framework for TSC. Specifically, we first construct a structured spatiotemporal graph to capture real-time traffic dynamics and spatial relationships among neighboring intersections, enabling the LLM to reason about complex traffic interactions. Moreover, we introduce a complexity-aware reasoning mechanism that dynamically adapts reasoning depth based on real-time traffic conditions, ensuring optimal computational efficiency without sacrificing decision quality. Besides, we propose a fine-tuning strategy that leverages iterative simulation-driven data collection and environmental feedback to build a lightweight LLM tailored for cooperative TSC. Extensive experiments on both synthetic and real-world datasets demonstrate that CoLLMLight outperforms state-of-the-art methods in diverse traffic scenarios, showcasing its effectiveness, scalability, and robustness.
Extracting structured labels from radiology reports has been employed to create vision models to simultaneously detect several types of abnormalities. However, existing works focus mainly on the chest region. Few works have been investigated on abdominal radiology reports due to more complex anatomy and a wider range of pathologies in the abdomen. We propose LEAVS (Large language model Extractor for Abdominal Vision Supervision). This labeler can annotate the certainty of presence and the urgency of seven types of abnormalities for nine abdominal organs on CT radiology reports. To ensure broad coverage, we chose abnormalities that encompass most of the finding types from CT reports. Our approach employs a specialized chain-of-thought prompting strategy for a locally-run LLM using sentence extraction and multiple-choice questions in a tree-based decision system. We demonstrate that the LLM can extract several abnormality types across abdominal organs with an average F1 score of 0.89, significantly outperforming competing labelers and humans. Additionally, we show that extraction of urgency labels achieved performance comparable to human annotations. Finally, we demonstrate that the abnormality labels contain valuable information for training a single vision model that classifies several organs as normal or abnormal. We release our code and structured annotations for a public CT dataset containing over 1,000 CT volumes.
Effective human-AI collaboration hinges not only on the AI agent's ability to follow explicit instructions but also on its capacity to navigate ambiguity, incompleteness, invalidity, and irrelevance in communication. Gricean conversational and inference norms facilitate collaboration by aligning unclear instructions with cooperative principles. We propose a normative framework that integrates Gricean norms and cognitive frameworks -- common ground, relevance theory, and theory of mind -- into large language model (LLM) based agents. The normative framework adopts the Gricean maxims of quantity, quality, relation, and manner, along with inference, as Gricean norms to interpret unclear instructions, which are: ambiguous, incomplete, invalid, or irrelevant. Within this framework, we introduce Lamoids, GPT-4 powered agents designed to collaborate with humans. To assess the influence of Gricean norms in human-AI collaboration, we evaluate two versions of a Lamoid: one with norms and one without. In our experiments, a Lamoid collaborates with a human to achieve shared goals in a grid world (Doors, Keys, and Gems) by interpreting both clear and unclear natural language instructions. Our results reveal that the Lamoid with Gricean norms achieves higher task accuracy and generates clearer, more accurate, and contextually relevant responses than the Lamoid without norms. This improvement stems from the normative framework, which enhances the agent's pragmatic reasoning, fostering effective human-AI collaboration and enabling context-aware communication in LLM-based agents.
Spatial reasoning is a fundamental capability of embodied agents and has garnered widespread attention in the field of multimodal large language models (MLLMs). In this work, we propose a novel benchmark, Open3DVQA, to comprehensively evaluate the spatial reasoning capacities of current state-of-the-art (SOTA) foundation models in open 3D space. Open3DVQA consists of 9k VQA samples, collected using an efficient semi-automated tool in a high-fidelity urban simulator. We evaluate several SOTA MLLMs across various aspects of spatial reasoning, such as relative and absolute spatial relationships, situational reasoning, and object-centric spatial attributes. Our results reveal that: 1) MLLMs perform better at answering questions regarding relative spatial relationships than absolute spatial relationships, 2) MLLMs demonstrate similar spatial reasoning abilities for both egocentric and allocentric perspectives, and 3) Fine-tuning large models significantly improves their performance across different spatial reasoning tasks. We believe that our open-source data collection tools and in-depth analyses will inspire further research on MLLM spatial reasoning capabilities. The benchmark is available at https://github.com/WeichenZh/Open3DVQA.