ICLR 2025 论文列表

Large language models (LLMs) augmented with retrieval exhibit robust performance and extensive versatility by incorporating external contexts. However, the input length grows linearly in the number of retrieved documents, causing a dramatic increase in latency. In this paper, we propose a novel paradigm named Sparse RAG, which seeks to cut computation costs through sparsity. Specifically, Sparse RAG encodes retrieved documents in parallel, which eliminates latency introduced by long-range attention of retrieved documents. Then, LLMs selectively decode the output by only attending to highly relevant caches auto-regressively, which are chosen via prompting LLMs with special control tokens. It is notable that Sparse RAG combines the assessment of each individual document and the generation of the response into a single process. The designed sparse mechanism in a RAG system can facilitate the reduction of the number of documents loaded during decoding for accelerating the inference of the RAG system. Additionally, filtering out undesirable contexts enhances the model’s focus on relevant context, inherently improving its generation quality. Evaluation results on four datasets show that Sparse RAG can be used to strike an optimal balance between generation quality and computational efficiency, demonstrating its generalizability across tasks.

Retrieval-augmented generation (RAG) is a key means to effectively enhance large language models (LLMs) in many knowledge-based tasks. However, existing RAG methods struggle with knowledge-intensive reasoning tasks, because useful information required to these tasks are badly scattered. This characteristic makes it difficult for existing RAG methods to accurately identify key information and perform global reasoning with such noisy augmentation. In this paper, motivated by the cognitive theories that humans convert raw information into various structured knowledge when tackling knowledge-intensive reasoning, we proposes a new framework, StructRAG, which can identify the optimal structure type for the task at hand, reconstruct original documents into this structured format, and infer answers based on the resulting structure. Extensive experiments across various knowledge-intensive tasks show that StructRAG achieves state-of-the-art performance, particularly excelling in challenging scenarios, demonstrating its potential as an effective solution for enhancing LLMs in complex real-world applications.

The scaling of inference computation has unlocked the potential of long-context large language models (LLMs) across diverse settings. For knowledge-intensive tasks, the increased compute is often allocated to incorporate more external knowledge. However, without effectively utilizing such knowledge, solely expanding context does not always enhance performance. In this work, we investigate inference scaling for retrieval augmented generation (RAG), exploring the combination of multiple strategies beyond simply increasing the quantity of knowledge, including in-context learning and iterative prompting. These strategies provide additional flexibility to scale test-time computation (e.g., by increasing retrieved documents or generation steps), thereby enhancing LLMs’ ability to effectively acquire and utilize contextual information. We address two key questions: (1) How does RAG performance benefit from the scaling of inference computation when optimally configured? (2) Can we predict the optimal test-time compute allocation for a given budget by modeling the relationship between RAG performance and inference parameters? Our observations reveal that increasing inference computation leads to nearly linear gains in RAG performance when optimally allocated, a relationship we describe as the inference scaling laws for RAG. Building on this, we further develop the computation allocation model to estimate RAG performance across different inference configurations. The model predicts optimal inference parameters under various computation constraints, which align closely with the experimental results. By applying these optimal configurations, we demonstrate that scaling inference compute on long-context LLMs achieves up to 58.9% gains on benchmark datasets compared to standard RAG.

Multimodal large language models (MLLMs) have recently shown great progress in text-rich image understanding, yet they still struggle with complex, multi-page visually-rich documents. Traditional methods using document parsers for retrieval-augmented generation suffer from performance and efficiency limitations, while directly presenting all pages to MLLMs leads to inefficiencies, especially with lengthy ones. In this work, we present a novel framework named **S**elf-**V**isual **R**etrieval-**A**ugmented **G**eneration (SV-RAG), which can broaden horizons of *any* MLLM to support long-document understanding. We demonstrate that **MLLMs themselves can be an effective multimodal retriever** to fetch relevant pages and then answer user questions based on these pages. SV-RAG is implemented with two specific MLLM adapters, one for evidence page retrieval and the other for question answering. Empirical results show state-of-the-art performance on public benchmarks, demonstrating the effectiveness of SV-RAG.

Large Language Models (LLMs) such as GPT-4 and Llama3 have significantly impacted various fields by enabling high-quality synthetic data generation and reducing dependence on expensive human-generated datasets. Despite this, challenges remain in the areas of generalization, controllability, diversity, and truthfulness within the existing generative frameworks. To address these challenges, this paper presents DataGen, a comprehensive LLM-powered framework designed to produce diverse, accurate, and highly controllable datasets. DataGen is adaptable, supporting all types of text datasets and enhancing the generative process through innovative mechanisms. To augment data diversity, DataGen incorporates an attribute-guided generation module and a group checking feature. For accuracy, it employs a code-based mathematical assessment for label verification alongside a retrieval-augmented generation technique for factual validation. The framework also allows for user-specified constraints, enabling customization of the data generation process to suit particular requirements. Extensive experiments demonstrate the superior quality of data generated by DataGen, and each module within DataGen plays a critical role in this enhancement. Additionally, DataGen is applied in two practical scenarios: benchmarking LLMs and data augmentation. The results indicate that DataGen effectively supports dynamic and evolving benchmarking and that data augmentation improves LLM capabilities in various domains, including agent-oriented abilities and reasoning skills.

Do diverse perspectives help students learn better? Multi-teacher knowledge distillation, which is a more effective technique than traditional single-teacher methods, supervises the student from different perspectives (i.e., teacher). While effective, multi-teacher, teacher ensemble, or teaching assistant-based approaches are computationally expensive and resource-intensive, as they require training multiple teacher networks. These concerns raise a question: can we supervise the student with diverse perspectives using only a single teacher? We, as the pioneer, demonstrate TeKAP, a novel teacher knowledge augmentation technique that generates multiple synthetic teacher knowledge by perturbing the knowledge of a single pretrained teacher i.e., Teacher Knowledge Augmentation via Perturbation, at both the feature and logit levels. These multiple augmented teachers simulate an ensemble of models together. The student model is trained on both the actual and augmented teacher knowledge, benefiting from the diversity of an ensemble without the need to train multiple teachers. TeKAP significantly reduces training time and computational resources, making it feasible for large-scale applications and easily manageable. Experimental results demonstrate that our proposed method helps existing state-of-the-art knowledge distillation techniques achieve better performance, highlighting its potential as a cost-effective alternative. The source code can be found in the supplementary.

Query-based sound separation (QSS) effectively isolate sound signals that match the content of a given query, enhancing the understanding of audio data. However, most existing QSS methods rely on a single modality for separation, lacking the ability to fully leverage homologous but heterogeneous information across multiple modalities for the same sound signal. To address this limitation, we introduce Omni-modal Sound Separation (**OmniSep**), a novel framework capable of isolating clean soundtracks based on omni-modal queries, encompassing both single-modal and multi-modal composed queries. Specifically, we introduce the **Query-Mixup** strategy, which blends query features from different modalities during training. This enables OmniSep to optimize multiple modalities concurrently, effectively bringing all modalities under a unified framework for sound separation. We further enhance this flexibility by allowing queries to influence sound separation positively or negatively, facilitating the retention or removal of specific sounds as desired. Finally, OmniSep employs a retrieval-augmented approach known as **Query-Aug**, which enables open-vocabulary sound separation. Experimental evaluations on MUSIC, VGGSOUND-CLEAN+, and MUSIC-CLEAN+ datasets demonstrate effectiveness of OmniSep, achieving state-of-the-art performance in text-, image-, and audio-queried sound separation tasks. For samples and further information, please visit the demo page at \url{https://omnisep.github.io/}.

In-context learning (ICL) and Retrieval-Augmented Generation (RAG) have gained attention for their ability to enhance LLMs' reasoning by incorporating external knowledge but suffer from limited contextual window size, leading to insufficient information injection. To this end, we propose a novel framework to automatically distill large volumes of offline data into interpretable first-order logic rules, which are injected into LLMs to boost their reasoning capabilities. Our method begins by formulating the search process relying on LLMs' commonsense, where LLMs automatically define head and body predicates. Then, we apply Monte Carlo Tree Search (MCTS) to address the combinational searching space and efficiently discover logic rules from data. The resulting logic rules are translated into natural language, allowing targeted knowledge injection and seamless integration into LLM prompts for LLM's downstream task reasoning. We evaluate our framework on public and private industrial tasks, including Natural Language Processing (NLP), time-series, decision-making, and industrial tasks, demonstrating its effectiveness in enhancing LLM's capability over diverse tasks.

The success of most existing cross-modal retrieval methods heavily relies on the assumption that the given queries follow the same distribution of the source domain. However, such an assumption is easily violated in real-world scenarios due to the complexity and diversity of queries, thus leading to the query shift problem. Specifically, query shift refers to the online query stream originating from the domain that follows a different distribution with the source one. In this paper, we observe that query shift would not only diminish the uniformity (namely, within-modality scatter) of the query modality but also amplify the gap between query and gallery modalities. Based on the observations, we propose a novel method dubbed Test-time adaptation for Cross-modal Retrieval (TCR). In brief, TCR employs a novel module to refine the query predictions (namely, retrieval results of the query) and a joint objective to prevent query shift from disturbing the common space, thus achieving online adaptation for the cross-modal retrieval models with query shift. Expensive experiments demonstrate the effectiveness of the proposed TCR against query shift. Code is available at https://github.com/XLearning-SCU/2025-ICLR-TCR.

Large language models (LLMs) have shown remarkable capabilities, but still struggle with processing extensive contexts, limiting their ability to maintain coherence and accuracy over long sequences. In contrast, the human brain excels at organising and retrieving episodic experiences across vast temporal scales, spanning a lifetime. In this work, we introduce EM-LLM, a novel approach that integrates key aspects of human episodic memory and event cognition into LLMs with no fine-tuning, enabling them to handle practically infinite context lengths while maintaining computational efficiency. EM-LLM organises sequences of tokens into coherent episodic events using a combination of Bayesian surprise and graph-theoretic boundary refinement in an online fashion. When needed, these events are retrieved through a two-stage memory process, combining similarity-based and temporally contiguous retrieval for efficient, human-inspired access to relevant information. Experiments on the LongBench and $\infty$-Bench benchmarks demonstrate EM-LLM's superior performance, consistently outperforming the state-of-the-art retrieval model InfLLM across various baseline LLMs. In addition, EM-LLM outperforms its popular counterpart, RAG, in a wide range of tasks, while requiring similar resources. Notably, EM-LLM's performance even surpasses full-context models in most tasks, while successfully performing retrieval across 10 million tokens -- a scale computationally infeasible for such models. Finally, our analysis reveals strong correlations between EM-LLM's event segmentation and human-perceived events, suggesting parallels between this artificial system and its biological counterpart, thereby offering a novel computational framework for exploring human memory mechanisms.

All text-based language problems can be reduced to either generation or embedding. Current models only perform well at one or the other. We introduce generative representational instruction tuning (GRIT) whereby a large language model is trained to handle both generative and embedding tasks by distinguishing between them through instructions. Compared to other open models, our resulting GritLM-7B is among the top models on the Massive Text Embedding Benchmark (MTEB) and outperforms various models up to its size on a range of generative tasks. By scaling up further, GritLM-8x7B achieves even stronger generative performance while still being among the best embedding models. Notably, we find that GRIT matches training on only generative or embedding data, thus we can unify both at no performance loss. Among other benefits, the unification via GRIT speeds up Retrieval-Augmented Generation (RAG) by > 60% for long documents, by no longer requiring separate retrieval and generation models. Models, code, etc. are freely available at https://github.com/ContextualAI/gritlm.

Large language models (LLMs) with long context windows have gained significant attention. However, the KV cache, stored to avoid re-computation, becomes a bottleneck. Various dynamic sparse or TopK-based attention approximation methods have been proposed to leverage the common insight that attention is sparse. In this paper, we first show that TopK attention itself suffers from quality degradation in certain downstream tasks because attention is not always as sparse as expected. Rather than selecting the keys and values with the highest attention scores, sampling with theoretical guarantees can provide a better estimation for attention output. To make the sampling-based approximation practical in LLM generation, we propose MagicPIG, a heterogeneous system based on Locality Sensitive Hashing (LSH). MagicPIG significantly reduces the workload of attention computation while preserving high accuracy for diverse tasks. MagicPIG stores the LSH hash tables and runs the attention computation on the CPU, which allows it to serve longer contexts and larger batch sizes with high approximation accuracy. MagicPIG can improve decoding throughput by up to $5\times$ across various GPU hardware and achieve 54ms decoding latency on a single RTX 4090 for Llama-3.1-8B-Instruct model with a context of 96k tokens.

Large Multimodal Models (LMMs) have made significant strides in visual question-answering for single images. Recent advancements like long-context LMMs have allowed them to ingest larger, or even multiple, images. However, the ability to process a large number of visual tokens does not guarantee effective retrieval and reasoning for multi-image question answering (MIQA), especially in real-world applications like photo album searches or satellite imagery analysis. In this work, we first assess the limitations of current benchmarks for long-context LMMs. We address these limitations by introducing a new vision-centric, long-context benchmark, "Visual Haystacks (VHs)". We comprehensively evaluate both open-source and proprietary models on VHs, and demonstrate that these models struggle when reasoning across potentially unrelated images, perform poorly on cross-image reasoning, as well as exhibit biases based on the placement of key information within the context window. Towards a solution, we introduce MIRAGE (Multi-Image Retrieval Augmented Generation), an open-source, lightweight visual-RAG framework that processes up to 10k images on a single 40G A100 GPU—far surpassing the 1k-image limit of contemporary models. MIRAGE demonstrates up to 13% performance improvement over existing open-source LMMs on VHs, sets a new state-of-the-art on the RetVQA multi-image QA benchmark, and achieves competitive performance on single-image QA with state-of-the-art LMMs. Our dataset, model, and code are available at: https://visual-haystacks.github.io.

Recent studies have shown that Large Language Models (LLMs) struggle to accurately retrieve information and maintain reasoning capabilities when processing long-context inputs. To address these limitations, we propose a finetuning approach utilizing a carefully designed synthetic dataset comprising numerical key-value retrieval tasks. Our experiments on models like GPT-3.5 Turbo and Mistral 7B demonstrate that finetuning LLMs on this dataset significantly improves LLMs' information retrieval and reasoning capabilities in longer-context settings. We present an analysis of the finetuned models, illustrating the transfer of skills from synthetic to real task evaluations (e.g., $10.5\%$ improvement on $20$ documents MDQA at position $10$ for GPT-3.5 Turbo). We also find that finetuned LLMs' performance on general benchmarks remains almost constant while LLMs finetuned on other baseline long-context augmentation data can encourage hallucination (e.g., on TriviaQA, Mistral 7B finetuned on our synthetic data cause no performance drop while other baseline data can cause a drop that ranges from $2.33\%$ to $6.19\%$). Our study highlights the potential of finetuning on synthetic data for improving the performance of LLMs on longer-context tasks.

Large language models (LLMs) have demonstrated significant potential in clinical decision support. Yet LLMs still suffer from hallucinations and lack fine-grained contextual medical knowledge, limiting their high-stake healthcare applications such as clinical diagnosis. Traditional retrieval-augmented generation (RAG) methods attempt to address these limitations but frequently retrieve sparse or irrelevant information, undermining prediction accuracy. We introduce KARE, a novel framework that integrates knowledge graph (KG) community-level retrieval with LLM reasoning to enhance healthcare predictions. KARE constructs a comprehensive multi-source KG by integrating biomedical databases, clinical literature, and LLM-generated insights, and organizes it using hierarchical graph community detection and summarization for precise and contextually relevant information retrieval. Our key innovations include: (1) a dense medical knowledge structuring approach enabling accurate retrieval of relevant information; (2) a dynamic knowledge retrieval mechanism that enriches patient contexts with focused, multi-faceted medical insights; and (3) a reasoning-enhanced prediction framework that leverages these enriched contexts to produce both accurate and interpretable clinical predictions. Extensive experiments demonstrate that KARE outperforms leading models by up to 10.8-15.0\% on MIMIC-III and 12.6-12.7\% on MIMIC-IV for mortality and readmission predictions. In addition to its impressive prediction accuracy, our framework leverages the reasoning capabilities of LLMs, enhancing the trustworthiness of clinical predictions.

Providing Large Language Models with relevant contextual knowledge at inference time has been shown to greatly improve the quality of their generations. This is often achieved by prepending informative passages of text, or 'contexts', retrieved from external knowledge bases to their input. However, processing additional contexts online incurs significant computation costs that scale with their length. State Space Models (SSMs) offer a promising solution by allowing a database of contexts to be mapped onto fixed-dimensional states from which to start the generation. A key challenge arises when attempting to leverage information present across multiple contexts, since there is no straightforward way to condition generation on multiple independent states in existing SSMs. To address this, we leverage a simple mathematical relation derived from SSM dynamics to compose multiple states into one that efficiently approximates the effect of concatenating raw context tokens. Since the temporal ordering of contexts can often be uninformative, we enforce permutation-invariance by efficiently averaging states obtained via our composition algorithm across all possible context orderings. We evaluate our resulting method on WikiText and MSMARCO in both zero-shot and fine-tuned settings, and show that we can match the strongest performing baseline while enjoying on average $5.4\times$ speedup.

We introduce Block-attention, an attention mechanism designed to address the increased inference latency and cost in Retrieval-Augmented Generation (RAG) scenarios. Traditional approaches often encode the entire context in an auto-regressive manner. Instead, Block-attention divides retrieved documents into discrete blocks, with each block independently calculating key-value (KV) states except for the final block. In RAG scenarios, by defining each passage as a block, Block-attention enables us to reuse the KV states of passages that have been seen before, thereby significantly reducing the latency and the computation overhead during inference. The implementation of Block-attention involves block segmentation, position re-encoding, and fine-tuning the LLM to adapt to the Block-attention mechanism. Experiments on 11 diverse benchmarks, including RAG, ICL, and general domains, demonstrate that after block fine-tuning, the Block-attention model not only achieves performance comparable to that of full-attention models, but can also seamlessly switch between the block and full attention modes without any performance loss. Notably, Block-attention significantly reduces the time to first token (TTFT) and floating point operations (FLOPs) to a very low level. It only takes 45 ms to output the first token for an input sequence with a total length of 32K. Compared to the full-attention models, the TTFT and corresponding FLOPs are reduced by 98.7\% and 99.8\%, respectively. Additionally, in Appendix A, we elaborate on how Block-attention is applied in Game AI scenario and the substantial potential benefits it entails. We strongly suggest researchers in the gaming field not to overlook this section.

When optimizing transportation systems, anticipating traffic flows is a central element. Yet, computing such traffic equilibria remains computationally expensive. Against this background, we introduce a novel combinatorial optimization augmented neural network pipeline that allows for fast and accurate traffic flow predictions. We propose WardropNet, a neural network that combines classical layers with a subsequent equilibrium layer: the first ones inform the latter by predicting the parameterization of the equilibrium problem's latency functions. Using supervised learning we minimize the difference between the actual traffic flow and the predicted output. We show how to leverage a Bregman divergence fitting the geometry of the equilibria, which allows for end-to-end learning. WardropNet outperforms pure learning-based approaches in predicting traffic equilibria for realistic and stylized traffic scenarios. On realistic scenarios, WardropNet improves on average for time-invariant predictions by up to 72\% and for time-variant predictions by up to 23\% over pure learning-based approaches.

The annotation (assigning structural chemical identities) of MS/MS spectra remains a significant challenge due to the enormous molecular diversity in biological samples and the limited scope of reference databases. Currently, the vast majority of spectral measurements remain in the "dark chemical space" without structural annotations. To improve annotation, we propose MADGEN (Mass-spec Attends to De Novo Molecular GENeration), a scaffold-based method for de novo molecular structure generation guided by mass spectrometry data. MADGEN operates in two stages: scaffold retrieval and spectra-conditioned molecular generation starting with the scaffold. In the first stage, given an MS/MS spectrum, we formulate scaffold retrieval as a ranking problem and employ contrastive learning to align mass spectra with candidate molecular scaffolds. In the second stage, starting from the retrieved scaffold, we employ the MS/MS spectrum to guide an attention-based generative model to generate the final molecule. Our approach constrains the molecular generation search space, reducing its complexity and improving generation accuracy. We evaluate MADGEN on three datasets (NIST23, CANOPUS, and MassSpecGym) and evaluate MADGEN's performance with a predictive scaffold retriever and with an oracle retriever. We demonstrate the effectiveness of using attention to integrate spectral information throughout the generation process to achieve strong results with the oracle retriever.