Ensuring the safety of large language model (LLM) applications is essential for developing trustworthy artificial intelligence. Current LLM safety benchmarks have two limitations. First, they focus solely on either discriminative or generative evaluation paradigms while ignoring their interconnection. Second, they rely on standardized inputs, overlooking the effects of widespread prompting techniques, such as system prompts, few-shot demonstrations, and chain-of-thought prompting. To overcome these issues, we developed SG-Bench, a novel benchmark to assess the generalization of LLM safety across various tasks and prompt types. This benchmark integrates both generative and discriminative evaluation tasks and includes extended data to examine the impact of prompt engineering and jailbreak on LLM safety. Our assessment of 3 advanced proprietary LLMs and 10 open-source LLMs with the benchmark reveals that most LLMs perform worse on discriminative tasks than generative ones, and are highly susceptible to prompts, indicating poor generalization in safety alignment. We also explain these findings quantitatively and qualitatively to provide insights for future research.
Developing algorithms to differentiate between machine-generated texts and human-written texts has garnered substantial attention in recent years. Existing methods in this direction typically concern an offline setting where a dataset containing a mix of real and machine-generated texts is given upfront, and the task is to determine whether each sample in the dataset is from a large language model (LLM) or a human. However, in many practical scenarios, sources such as news websites, social media accounts, or on other forums publish content in a streaming fashion. Therefore, in this online scenario, how to quickly and accurately determine whether the source is an LLM with strong statistical guarantees is crucial for these media or platforms to function effectively and prevent the spread of misinformation and other potential misuse of LLMs. To tackle the problem of online detection, we develop an algorithm based on the techniques of sequential hypothesis testing by betting that not only builds upon and complements existing offline detection techniques but also enjoys statistical guarantees, which include a controlled false positive rate and the expected time to correctly identify a source as an LLM. Experiments were conducted to demonstrate the effectiveness of our method.
Generative models such as Large Language Models (LLM) and Multimodal Large Language models (MLLMs) trained on massive web corpora can memorize and disclose individuals' confidential and private data, raising legal and ethical concerns. While many previous works have addressed this issue in LLM via machine unlearning, it remains largely unexplored for MLLMs. To tackle this challenge, we introduce Multimodal Large Language Model Unlearning Benchmark (MLLMU-Bench), a novel benchmark aimed at advancing the understanding of multimodal machine unlearning. MLLMU-Bench consists of 500 fictitious profiles and 153 profiles for public celebrities, each profile feature over 14 customized question-answer pairs, evaluated from both multimodal (image+text) and unimodal (text) perspectives. The benchmark is divided into four sets to assess unlearning algorithms in terms of efficacy, generalizability, and model utility. Finally, we provide baseline results using existing generative model unlearning algorithms. Surprisingly, our experiments show that unimodal unlearning algorithms excel in generation and cloze tasks, while multimodal unlearning approaches perform better in classification tasks with multimodal inputs.
Although large language models (LLMs) have demonstrated their strong intelligence ability, the high demand for computation and storage hinders their practical application. To this end, many model compression techniques are proposed to increase the efficiency of LLMs. However, current researches only validate their methods on limited models, datasets, metrics, etc, and still lack a comprehensive evaluation under more general scenarios. So it is still a question of which model compression approach we should use under a specific case. To mitigate this gap, we present the Large Language Model Compression Benchmark (LLMCBench), a rigorously designed benchmark with an in-depth analysis for LLM compression algorithms. We first analyze the actual model production requirements and carefully design evaluation tracks and metrics. Then, we conduct extensive experiments and comparison using multiple mainstream LLM compression approaches. Finally, we perform an in-depth analysis based on the evaluation and provide useful insight for LLM compression design. We hope our LLMCBench can contribute insightful suggestions for LLM compression algorithm design and serve as a foundation for future research. Our code is available at https://github.com/AboveParadise/LLMCBench.
Large language models (LLMs) have not only revolutionized the field of natural language processing (NLP) but also have the potential to bring a paradigm shift in many other fields due to their remarkable abilities of language understanding, as well as impressive generalization capabilities and reasoning skills. As a result, recent studies have actively attempted to harness the power of LLMs to improve recommender systems, and it is imperative to thoroughly review the recent advances and challenges of LLM-based recommender systems. Unlike existing work, this survey does not merely analyze the classifications of LLM-based recommendation systems according to the technical framework of LLMs. Instead, it investigates how LLMs can better serve recommendation tasks from the perspective of the recommender system community, thus enhancing the integration of large language models into the research of recommender system and its practical application. In addition, the long-standing gap between academic research and industrial applications related to recommender systems has not been well discussed, especially in the era of large language models. In this review, we introduce a novel taxonomy that originates from the intrinsic essence of recommendation, delving into the application of large language model-based recommendation systems and their industrial implementation. Specifically, we propose a three-tier structure that more accurately reflects the developmental progression of recommendation systems from research to practical implementation, including representing and understanding, scheming and utilizing, and industrial deployment. Furthermore, we discuss critical challenges and opportunities in this emerging field. A more up-to-date version of the papers is maintained at: https://github.com/jindongli-Ai/Next-Generation-LLM-based-Recommender-Systems-Survey.
The synthesis of inorganic crystalline materials is essential for modern technology, especially in quantum materials development. However, designing efficient synthesis workflows remains a significant challenge due to the precise experimental conditions and extensive trial and error. Here, we present a framework using large language models (LLMs) to predict synthesis pathways for inorganic materials, including quantum materials. Our framework contains three models: LHS2RHS, predicting products from reactants; RHS2LHS, predicting reactants from products; and TGT2CEQ, generating full chemical equations for target compounds. Fine-tuned on a text-mined synthesis database, our model raises accuracy from under 40% with pretrained models, to under 80% using conventional fine-tuning, and further to around 90% with our proposed generalized Tanimoto similarity, while maintaining robust to additional synthesis steps. Our model further demonstrates comparable performance across materials with varying degrees of quantumness quantified using quantum weight, indicating that LLMs offer a powerful tool to predict balanced chemical equations for quantum materials discovery.
Large language model-based agents are increasingly applied in the recommendation field due to their extensive knowledge and strong planning capabilities. While prior research has primarily focused on enhancing either the recommendation agent or the user agent individually, the collaborative interaction between the two has often been overlooked. Towards this research gap, we propose a novel framework that emphasizes the feedback loop process to facilitate the collaboration between the recommendation agent and the user agent. Specifically, the recommendation agent refines its understanding of user preferences by analyzing the feedback from the user agent on the item recommendation. Conversely, the user agent further identifies potential user interests based on the items and recommendation reasons provided by the recommendation agent. This iterative process enhances the ability of both agents to infer user behaviors, enabling more effective item recommendations and more accurate user simulations. Extensive experiments on three datasets demonstrate the effectiveness of the agentic feedback loop: the agentic feedback loop yields an average improvement of 11.52% over the single recommendation agent and 21.12% over the single user agent. Furthermore, the results show that the agentic feedback loop does not exacerbate popularity or position bias, which are typically amplified by the real-world feedback loop, highlighting its robustness. The source code is available at https://github.com/Lanyu0303/AFL.
Our interest is in constructing interactive systems involving a human-expert interacting with a machine learning engine on data analysis tasks. This is of relevance when addressing complex problems arising in areas of science, the environment, medicine and so on, which are not immediately amenable to the usual methods of statistical or mathematical modelling. In such situations, it is possible that harnessing human expertise and creativity to modern machine-learning capabilities of identifying patterns by constructing new internal representations of the data may provide some insight to possible solutions. In this paper, we examine the implementation of an abstract protocol developed for interaction between agents, each capable of constructing predictions and explanations. The \PXP protocol, described in [12] is motivated by the notion of ''two-way intelligibility'' and is specified using a pair of communicating finite-state machines. While the formalisation allows the authors to prove several properties about the protocol, no implementation was presented. Here, we address this shortcoming for the case in which one of the agents acts as a ''generator'' using a large language model (LLM) and the other is an agent that acts as a ''tester'' using either a human-expert, or a proxy for a human-expert (for example, a database compiled using human-expertise). We believe these use-cases will be a widely applicable form of interaction for problems of the kind mentioned above. We present an algorithmic description of general-purpose implementation, and conduct preliminary experiments on its use in two different areas (radiology and drug-discovery). The experimental results provide early evidence in support of the protocol's capability of capturing one- and two-way intelligibility in human-LLM in the manner proposed in [12].
Trajectory modeling, which includes research on trajectory data pattern mining and future prediction, has widespread applications in areas such as life services, urban transportation, and public administration. Numerous methods have been proposed to address specific problems within trajectory modelling. However, due to the heterogeneity of data and the diversity of trajectory tasks, achieving unified trajectory modelling remains an important yet challenging task. In this paper, we propose TrajAgent, a large language model-based agentic framework, to unify various trajectory modelling tasks. In TrajAgent, we first develop UniEnv, an execution environment with a unified data and model interface, to support the execution and training of various models. Building on UniEnv, we introduce TAgent, an agentic workflow designed for automatic trajectory modelling across various trajectory tasks. Specifically, we design AutOpt, a systematic optimization module within TAgent, to further improve the performance of the integrated model. With diverse trajectory tasks input in natural language, TrajAgent automatically generates competitive results via training and executing appropriate models. Extensive experiments on four tasks using four real-world datasets demonstrate the effectiveness of TrajAgent in unified trajectory modelling, achieving an average performance improvement of 15.43% over baseline methods.
This study introduces SLLMBO, an innovative framework leveraging large language models (LLMs) for hyperparameter optimization (HPO), incorporating dynamic search space adaptability, enhanced parameter space exploitation, and a novel LLM-tree-structured parzen estimator (LLM-TPE) sampler. By addressing limitations in recent fully LLM-based methods and traditional bayesian optimization (BO), SLLMBO achieves more robust optimization. This comprehensive benchmarking evaluates multiple LLMs, including GPT-3.5-Turbo, GPT-4o, Claude-Sonnet-3.5, and Gemini-1.5-Flash, extending prior work and establishing SLLMBO as the first framework to benchmark a diverse set of LLMs for HPO. By integrating LLMs' established strengths in parameter initialization with the exploitation abilities demonstrated in this study, alongside TPE's exploration capabilities, the LLM-TPE sampler achieves a balanced exploration-exploitation trade-off, reduces API costs, and mitigates premature early stoppings for more effective parameter searches. Across 14 tabular tasks in classification and regression, the LLM-TPE sampler outperformed fully LLM-based methods and achieved superior results over BO methods in 9 tasks. Testing early stopping in budget-constrained scenarios demonstrated competitive performance, indicating that LLM-based methods generally benefit from extended iterations for optimal results. This work lays the foundation for future research exploring open-source LLMs, reproducibility of LLM results in HPO, and benchmarking SLLMBO on complex datasets, such as image classification, segmentation, and machine translation.
Software engineers operating in complex and dynamic environments must continuously adapt to evolving requirements, learn iteratively from experience, and reconsider their approaches based on new insights. However, current large language model (LLM)-based software agents often rely on rigid processes and tend to repeat ineffective actions without the capacity to evaluate their performance or adapt their strategies over time. To address these challenges, we propose SWE-Search, a multi-agent framework that integrates Monte Carlo Tree Search (MCTS) with a self-improvement mechanism to enhance software agents' performance on repository-level software tasks. SWE-Search extends traditional MCTS by incorporating a hybrid value function that leverages LLMs for both numerical value estimation and qualitative evaluation. This enables self-feedback loops where agents iteratively refine their strategies based on both quantitative numerical evaluations and qualitative natural language assessments of pursued trajectories. The framework includes a SWE-Agent for adaptive exploration, a Value Agent for iterative feedback, and a Discriminator Agent that facilitates multi-agent debate for collaborative decision-making. Applied to the SWE-bench benchmark, our approach demonstrates a 23% relative improvement in performance across five models compared to standard open-source agents without MCTS. Our analysis reveals how performance scales with increased search depth and identifies key factors that facilitate effective self-evaluation in software agents. This work highlights the potential of self-evaluation driven search techniques to enhance agent reasoning and planning in complex, dynamic software engineering environments.
Large Language Models (LLMs) have demonstrated strong reasoning abilities, making them suitable for complex tasks such as graph computation. Traditional reasoning steps paradigm for graph problems is hindered by unverifiable steps, limited long-term reasoning, and poor generalization to graph variations. To overcome these limitations, we introduce GCoder, a code-based LLM designed to enhance problem-solving in generalized graph computation problems. Our method involves constructing an extensive training dataset, GraphWild, featuring diverse graph formats and algorithms. We employ a multi-stage training process, including Supervised Fine-Tuning (SFT) and Reinforcement Learning from Compiler Feedback (RLCF), to refine model capabilities. For unseen tasks, a hybrid retrieval technique is used to augment performance. Experiments demonstrate that GCoder outperforms GPT-4o, with an average accuracy improvement of 16.42% across various graph computational problems. Furthermore, GCoder efficiently manages large-scale graphs with millions of nodes and diverse input formats, overcoming the limitations of previous models focused on the reasoning steps paradigm. This advancement paves the way for more intuitive and effective graph problem-solving using LLMs. Code and data are available at here: https://github.com/Bklight999/WWW25-GCoder/tree/master.
FP8 training has emerged as a promising method for improving training efficiency. Existing frameworks accelerate training by applying FP8 computation to linear layers while leaving optimizer states and activations in higher precision, which fails to fully optimize memory usage. This paper introduces COAT (Compressing Optimizer States and Activations for FP8 Training), a novel FP8 training framework designed to significantly reduce memory footprint when training large models. COAT addresses current limitations through two key innovations: (1) Dynamic Range Expansion, which aligns optimizer state distributions more closely with the FP8 representation range, thereby reducing quantization error, and (2) Mixed-Granularity Activation Quantization, which optimizes activation memory using a combination of per-tensor and per-group quantization strategies. Experiments demonstrate that COAT effectively reduces end-to-end training memory footprint by 1.54x compared to BF16 while achieving nearly lossless performance across various tasks, such as Large Language Model pretraining and fine-tuning and Vision Language Model training. COAT also achieves a 1.43x end-to-end training speedup compared to BF16, performing on par with or surpassing TransformerEngine's speedup. COAT enables efficient full-parameter training of large models on fewer GPUs, and facilitates doubling the batch size in distributed training settings, providing a practical solution for scaling large-scale model training. The code is available at https://github.com/NVlabs/COAT.
The dominant paradigm for RLHF is online and on-policy RL: synchronously generating from the large language model (LLM) policy, labelling with a reward model, and learning using feedback on the LLM's own outputs. While performant, this paradigm is computationally inefficient. Inspired by classical deep RL literature, we propose separating generation and learning in RLHF. This enables asynchronous generation of new samples while simultaneously training on old samples, leading to faster training and more compute-optimal scaling. However, asynchronous training relies on an underexplored regime, online but off-policy RLHF: learning on samples from previous iterations of our model. To understand the challenges in this regime, we investigate a fundamental question: how much off-policyness can we tolerate for asynchronous training to speed up learning but maintain performance? Among several RLHF algorithms we tested, we find that online DPO is most robust to off-policy data, and robustness increases with the scale of the policy model. We study further compute optimizations for asynchronous RLHF but find that they come at a performance cost, giving rise to a trade-off. Finally, we verify the scalability of asynchronous RLHF by training LLaMA 3.1 8B on an instruction-following task 40% faster than a synchronous run while matching final performance.
Extending Large Language Models (LLMs) to advanced applications requires reliable structured output generation. Existing methods which often rely on rigid JSON schemas, can lead to unreliable outputs, diminished reasoning capabilities, and increased computational overhead, limiting LLMs' adaptability for complex tasks. We introduce Meaning Typed Prompting (MTP), a technique for efficient structured output generation that integrates types, meanings, and abstractions, such as variables and classes, into the prompting process. By utilizing expressive type definitions, MTP enhances output clarity and reduces dependence on complex abstractions, simplifying development, and improving implementation efficiency. This enables LLMs to understand relationships and generate structured data more effectively. Empirical evaluations on multiple benchmarks demonstrate that MTP outperforms existing frameworks in accuracy, reliability, consistency, and token efficiency. We present Semantix, a framework that implements MTP, providing practical insights into its application.
Training data attribution (TDA) methods aim to attribute model outputs back to specific training examples, and the application of these methods to large language model (LLM) outputs could significantly advance model transparency and data curation. However, it has been challenging to date to apply these methods to the full scale of LLM pretraining. In this paper, we refine existing gradient-based methods to work effectively at scale, allowing us to retrieve influential examples for an 8B-parameter language model from a pretraining corpus of over 160B tokens with no need for subsampling or pre-filtering. Our method combines several techniques, including optimizer state correction, a task-specific Hessian approximation, and normalized encodings, which we find to be critical for performance at scale. In quantitative evaluations on a fact tracing task, our method performs best at identifying examples that influence model predictions, but classical, model-agnostic retrieval methods such as BM25 still perform better at finding passages which explicitly contain relevant facts. These results demonstrate a misalignment between factual *attribution* and causal *influence*. With increasing model size and training tokens, we find that influence more closely aligns with factual attribution. Finally, we examine different types of examples identified as influential by our method, finding that while many directly entail a particular fact, others support the same output by reinforcing priors on relation types, common entities, and names. We release our prompt set and model outputs, along with a web-based visualization tool to explore influential examples for factual predictions, commonsense reasoning, arithmetic, and open-ended generation for an 8B-parameter LLM.
The correct specification of reward models is a well-known challenge in reinforcement learning. Hand-crafted reward functions often lead to inefficient or suboptimal policies and may not be aligned with user values. Reinforcement learning from human feedback is a successful technique that can mitigate such issues, however, the collection of human feedback can be laborious. Recent works have solicited feedback from pre-trained large language models rather than humans to reduce or eliminate human effort, however, these approaches yield poor performance in the presence of hallucination and other errors. This paper studies the advantages and limitations of reinforcement learning from large language model feedback and proposes a simple yet effective method for soliciting and applying feedback as a potential-based shaping function. We theoretically show that inconsistent rankings, which approximate ranking errors, lead to uninformative rewards with our approach. Our method empirically improves convergence speed and policy returns over commonly used baselines even with significant ranking errors, and eliminates the need for complex post-processing of reward functions.
Graphs are widely used for modeling relational data in real-world scenarios, such as social networks and urban computing. Existing LLM-based graph analysis approaches either integrate graph neural networks (GNNs) for specific machine learning tasks, limiting their transferability, or rely solely on LLMs' internal reasoning ability, resulting in suboptimal performance. To address these limitations, we take advantage of recent advances in LLM-based agents, which have shown capabilities of utilizing external knowledge or tools for problem solving. By simulating human problem-solving strategies such as analogy and collaboration, we propose a multi-agent system based on LLMs named GraphTeam, for graph analysis. GraphTeam consists of five LLM-based agents from three modules, and the agents with different specialities can collaborate with each other to address complex problems. Specifically, (1) input-output normalization module: the question agent extracts and refines four key arguments from the original question, facilitating the problem understanding, and the answer agent organizes the results to meet the output requirement; (2) external knowledge retrieval module: we first build a knowledge base consisting of relevant documentation and experience information, and then the search agent retrieves the most relevant entries for each question. (3) problem-solving module: given the retrieved information from search agent, the coding agent uses established algorithms via programming to generate solutions, and in case the coding agent does not work, the reasoning agent will directly compute the results without programming. Extensive experiments on six graph analysis benchmarks demonstrate that GraphTeam achieves state-of-the-art performance with an average 25.85% improvement over the best baseline in terms of accuracy. The code and data are available at https://github.com/BUPT-GAMMA/GraphTeam.
Full-duplex spoken dialogue systems significantly surpass traditional turn-based dialogue systems, as they allow simultaneous bidirectional communication, closely mirroring human-human interactions. However, achieving low latency and natural interactions in full-duplex dialogue systems remains a significant challenge, especially considering human conversation dynamics such as interruptions, backchannels, and overlapping speech. In this paper, we introduce a novel End-to-End GPT-based model OmniFlatten for full-duplex conversation, capable of effectively modeling the complex behaviors inherent to natural conversations with low latency. To achieve full-duplex conversation capabilities, we propose a multi-stage post-training scheme that progressively adapts a text large language model (LLM) backbone into a speech-text dialogue LLM, capable of generating text and speech in real time, without modifying the architecture of the backbone LLM. The training process comprises three stages: modality alignment, half-duplex dialogue learning, and full-duplex dialogue learning. In all training stages, we standardize the data using a flattening operation, which enables unifying the training methods and the GPT backbone across different modalities and tasks. Our approach offers a simple modeling technique and a promising research direction for developing efficient and natural end-to-end full-duplex spoken dialogue systems. Audio samples of dialogues generated by OmniFlatten can be found at this web site (https://omniflatten.github.io/).
There are already many DNA large language models, but most of them still follow traditional uses, such as extracting sequence features for classification tasks. More innovative applications of large language models, such as prompt engineering, RAG, and zero-shot or few-shot prediction, remain challenging for DNA-based models. The key issue lies in the fact that DNA models and human natural language models are entirely separate; however, techniques like prompt engineering require the use of natural language, thereby significantly limiting the application of DNA large language models. This paper introduces a pre-trained model trained on the GPT-2 network, combining DNA sequences and English text, and uses a unified BPE tokenization method. We then convert classification and other downstream tasks into Alpaca format instruction data, and perform instruction fine-tuning on this pre-trained model to create a fine-tuned model capable of handling multiple tasks. The model has demonstrated its effectiveness in DNA related zero-shot prediction and multitask application. This research provides a highly promising direction for building a unified DNA sequence task framework.