Vision-language models (VLMs) have advanced human-AI interaction but struggle with cultural understanding, often misinterpreting symbols, gestures, and artifacts due to biases in predominantly Western-centric training data. In this paper, we construct CultureVerse, a large-scale multimodal benchmark covering 19,682 cultural concepts, 188 countries/regions, 15 cultural topics, and 3 question types, with the aim of characterizing and improving VLMs’ multicultural understanding capabilities. Then, we propose CultureVLM, a series of VLMs fine-tuned on our dataset to achieve significant performance improvement in cultural understanding. Our evaluation of 16 models reveals significant disparities, with a stronger performance in Western concepts and weaker results in African and Asian contexts. Fine-tuning on our CultureVerse training set enhances cultural perception, demonstrating cross-cultural, cross-continent, and cross-dataset generalization without sacrificing performance on models’ general VLM benchmarks. We further present insights on cultural generalization and forgetting. We hope that this work could lay the foundation for more equitable and culturally aware multimodal AI systems.

Large language models (LLMs) have made significant progress in various NLP tasks, especially with the rise of long-context models that handle extended inputs. However, the growing size of the key-value (KV) cache in Transformer architectures imposes heavy memory demands during decoding, becoming a major bottleneck. Existing sparse attention methods aim to address this issue but face two main limitations: (1) they introduce high overhead when retrieving important tokens for attention, and (2) they often fail to significantly reduce memory usage in practice. This paper presents Seer, a simple yet effective system for fast and accurate LLM decoding on long-context sequences. Seer leverages semantic space representations to manage the KV cache more efficiently. It dynamically estimates cluster-level importance to reduce both computation and memory access, while preserving output quality. Through offline training, Seer learns semantic centroids to enable fast token importance estimation. To further improve efficiency, it integrates product quantization with sparse attention for KV cache compression and uses an asymmetric distance lookup table to reduce computational cost. Evaluation on a diverse set of LLMs and tasks shows that Seer reduces theoretical attention bandwidth by 40× with minimal accuracy loss. Seer achieves a 5.57× attention speedup and a 2× End-To-End speedup over full attention at 32K sequence length. Our code is available at https://github.com/Acaciasama/Seer.

Fine-tuning on agent-environment interaction trajectory data holds significant promise for surfacing generalized agent capabilities in open-source large language models (LLMs). In this work, we introduce AgentBank, by far the largest trajectory tuning data collection featuring more than 50k diverse high-quality interaction trajectories which comprises 16 tasks covering five distinct agent skill dimensions. Leveraging a novel annotation pipeline, we are able to scale the annotated trajectories and generate a trajectory dataset with minimized difficulty bias. Furthermore, we fine-tune LLMs on AgentBank to get a series of agent models, Samoyed. Our comparative experiments demonstrate the effectiveness of scaling the interaction trajectory data to acquire generalized agent capabilities. Additional studies also reveal some key observations regarding trajectory tuning and agent skill generalization.

Large language model agents have exhibited exceptional performance across a range of complex interactive tasks. Recent approaches have utilized tuning with expert trajectories to enhance agent performance, yet they primarily concentrate on outcome rewards, which may lead to errors or suboptimal actions due to the absence of process supervision signals. In this paper, we introduce the Iterative step-level Process Refinement (IPR) framework, which provides detailed step-by-step guidance to enhance agent training. Specifically, we adopt the Monte Carlo method to estimate step-level rewards. During each iteration, the agent explores along the expert trajectory and generates new actions. These actions are then evaluated against the corresponding step of expert trajectory using step-level rewards. Such comparison helps identify discrepancies, yielding contrastive action pairs that serve as training data for the agent. Our experiments on three complex agent tasks demonstrate that our framework outperforms a variety of strong baselines. Moreover, our analytical finds highlight the effectiveness of IPR in augmenting action efficiency and its applicability to diverse models.

Retrieval Augmented Generation (RAG) has been a powerful tool for Large Language Models (LLMs) to efficiently process overly lengthy contexts. However, recent LLMs like Gemini-1.5 and GPT-4 show exceptional capabilities to understand long contexts directly. We conduct a comprehensive comparison between RAG and long-context (LC) LLMs, aiming to leverage the strengths of both. We benchmark RAG and LC across various public datasets using three latest LLMs. Results reveal that when resourced sufficiently, LC consistently outperforms RAG in terms of average performance. However, RAG’s significantly lower cost remains a distinct advantage. Based on this observation, we propose Self-Route, a simple yet effective method that routes queries to RAG or LC based on model self-reflection. Self-Route significantly reduces the computation cost while maintaining a comparable performance to LC. Our findings provide a guideline for long-context applications of LLMs using RAG and LC.

Large Language Models (LLMs) have demonstrated superior results across a wide range of tasks, and Retrieval-augmented Generation (RAG) is an effective way to enhance the performance by locating relevant information and placing it into the context window of the LLM. However, the relationship between retrievers and LLMs in a RAG is still under-investigated. Most existing work treats the retriever and the LLM as independent components and leaves a gap between retrieving human-”friendly” information and assembling a LLM-”friendly” context. In this work, we examine a novel bridge mechanism. We validate the ranking and selection assumptions of retrievers in the context of RAG and propose a framework that chains together supervised and reinforcement learning to train a bridge model that optimizes the connection between the retriever and the LLM. Empirical results demonstrate the effectiveness of our method in both question-answering and personalized generation tasks.

Role-playing agents (RPAs), powered by large language models, have emerged as a flourishing field of applications. However, a key challenge lies in assessing whether RPAs accurately reproduce the personas of target characters, namely their character fidelity. Existing methods mainly focus on the knowledge and linguistic patterns of characters. This paper, instead, introduces a novel perspective to evaluate the personality fidelity of RPAs with psychological scales. Overcoming drawbacks of previous self-report assessments on RPAs, we propose InCharacter, namely Interviewing Character agents for personality tests. Experiments include various types of RPAs and LLMs, covering 32 distinct characters on 14 widely used psychological scales. The results validate the effectiveness of InCharacter in measuring RPA personalities. Then, with InCharacter, we show that state-of-the-art RPAs exhibit personalities highly aligned with the human-perceived personalities of the characters, achieving an accuracy up to 80.7%.

Further pre-training language models on in-domain data (domain-adaptive pre-training, DAPT) or task-relevant data (task-adaptive pre-training, TAPT) before fine-tuning has been shown to improve downstream tasks’ performances. However, in task-oriented dialog modeling, we observe that further pre-training MLM does not always boost the performance on a downstream task. We find that DAPT is beneficial in the low-resource setting, but as the fine-tuning data size grows, DAPT becomes less beneficial or even useless, and scaling the size of DAPT data does not help. Through Representational Similarity Analysis, we conclude that more data for fine-tuning yields greater change of the model’s representations and thus reduces the influence of initialization.

Most of the existing Knowledge-based Question Answering (KBQA) methods first learn to map the given question to a query graph, and then convert the graph to an executable query to find the answer. The query graph is typically expanded progressively from the topic entity based on a sequence prediction model. In this paper, we propose a new solution to query graph generation that works in the opposite manner: we start with the entire knowledge base and gradually shrink it to the desired query graph. This approach improves both the efficiency and the accuracy of query graph generation, especially for complex multi-hop questions. Experimental results show that our method achieves state-of-the-art performance on ComplexWebQuestion (CWQ) dataset.

Most language understanding models in task-oriented dialog systems are trained on a small amount of annotated training data, and evaluated in a small set from the same distribution. However, these models can lead to system failure or undesirable output when being exposed to natural language perturbation or variation in practice. In this paper, we conduct comprehensive evaluation and analysis with respect to the robustness of natural language understanding models, and introduce three important aspects related to language understanding in real-world dialog systems, namely, language variety, speech characteristics, and noise perturbation. We propose a model-agnostic toolkit LAUG to approximate natural language perturbations for testing the robustness issues in task-oriented dialog. Four data augmentation approaches covering the three aspects are assembled in LAUG, which reveals critical robustness issues in state-of-the-art models. The augmented dataset through LAUG can be used to facilitate future research on the robustness testing of language understanding in task-oriented dialog.

We present an adaptation of RNN sequence models to the problem of multi-label classification for text, where the target is a set of labels, not a sequence. Previous such RNN models define probabilities for sequences but not for sets; attempts to obtain a set probability are after-thoughts of the network design, including pre-specifying the label order, or relating the sequence probability to the set probability in ad hoc ways. Our formulation is derived from a principled notion of set probability, as the sum of probabilities of corresponding permutation sequences for the set. We provide a new training objective that maximizes this set probability, and a new prediction objective that finds the most probable set on a test document. These new objectives are theoretically appealing because they give the RNN model freedom to discover the best label order, which often is the natural one (but different among documents). We develop efficient procedures to tackle the computation difficulties involved in training and prediction. Experiments on benchmark datasets demonstrate that we outperform state-of-the-art methods for this task.