In this work, we investigate a typical scenario in code generation where a developer edits existing code in real time and requests a code assistant, e.g., a large language model, to re-predict the next token or next line on the fly. Naively, the LLM needs to re-encode the entire KV cache to provide an accurate prediction. However, this process is computationally expensive, especially when the sequence length is long. Simply encoding the edited subsequence and integrating it to the original KV cache meets the temporal confusion problem, leading to significantly worse performance. We address this efficiency and accuracy trade-off by introducing \underline{\textbf{Positional \textbf{I}ntegrity \textbf{E}ncoding} (PIE). Building upon the rotary positional encoding, PIE first removes the rotary matrices in the Key cache that introduce temporal confusion and then reapplies the correct rotary matrices. This process ensures that positional relationships between tokens are correct and requires only a single round of matrix multiplication. We validate the effectiveness of PIE through extensive experiments on the RepoBench-C-8k dataset, utilizing DeepSeek-Coder models with 1.3B, 6.7B, and 33B parameters. Our evaluation includes three real-world coding tasks: code insertion, code deletion, and multi-place code editing. Results demonstrate that PIE reduces computational overhead by over 85% compared to the standard full recomputation approach across all model sizes and tasks while well approximating the model performance.

Multimodal foundation models that can holistically process text alongside images, video, audio, and other sensory modalities are increasingly used in a variety of real-world applications. However, it is challenging to characterize and study progress in multimodal foundation models, given the range of possible modeling decisions, tasks, and domains. In this paper, we introduce Holistic Evaluation of Multimodal Models (HEMM) to systematically evaluate the capabilities of multimodal foundation models across a set of 3 dimensions: basic skills, information flow, and real-world use cases. Basic multimodal skills are internal abilities required to solve problems, such as learning interactions across modalities, fine-grained alignment, multi-step reasoning, and the ability to handle external knowledge. Information flow studies how multimodal content changes during a task through querying, translation, editing, and fusion. Use cases span domain-specific challenges introduced in real-world multimedia, affective computing, natural sciences, healthcare, and human-computer interaction applications. Through comprehensive experiments across the 30 tasks in HEMM, we (1) identify key dataset dimensions (e.g., basic skills, information flows, and use cases) that pose challenges to today's models, and (2) distill performance trends regarding how different modeling dimensions (e.g., scale, pre-training data, multimodal alignment, pre-training, and instruction tuning objectives) influence performance. Our conclusions regarding challenging multimodal interactions, use cases, and tasks requiring reasoning and external knowledge, the benefits of data and model scale, and the impacts of instruction tuning yield actionable insights for future work in multimodal foundation models.

Large language models (LLMs) are used for long-form question answering (LFQA), which requires them to generate paragraph-length answers to complex questions. While LFQA has been well-studied in English, this research has not been extended to other languages. To bridge this gap, we introduce CaLMQA, a collection of 1.5K complex culturally specific questions spanning 23 languages and 51 culturally agnostic questions translated from English into 22 other languages. We define culturally specific questions as those uniquely or more likely to be asked by people from cultures associated with the question's language. We collect naturally-occurring questions from community web forums and hire native speakers to write questions to cover under-resourced, rarely-studied languages such as Fijian and Kirundi. Our dataset contains diverse, complex questions that reflect cultural topics (e.g. traditions, laws, news) and the language usage of native speakers. We automatically evaluate a suite of open- and closed-source models on CaLMQA by detecting incorrect language and token repetitions in answers, and observe that the quality of LLM-generated answers degrades significantly for some low-resource languages. Lastly, we perform human evaluation on a subset of models and languages. Manual evaluation reveals that model performance is significantly worse for culturally specific questions than for culturally agnostic questions. Our findings highlight the need for further research in non-English LFQA and provide an evaluation framework.

End-to-end Task-Oriented Dialog (TOD) systems typically require extensive training datasets to perform well. In contrast, large language model (LLM) based TOD systems can excel even with limited data due to their ability to learn tasks through in-context exemplars. However, these models lack alignment with the style of responses in training data and often generate comprehensive responses, making it difficult for users to grasp the information quickly. In response, we propose SyncTOD that synergizes LLMs with task-specific hints to improve alignment in low-data settings. SyncTOD employs small auxiliary models to provide hints and select exemplars for in-context prompts. With ChatGPT, SyncTOD achieves superior performance compared to LLM-based baselines and SoTA models in low-data settings, while retaining competitive performance in full-data settings.

Integrating cognitive ergonomics with LLMs is essential for enhancing safety, reliability, and user satisfaction in human-AI interactions. Current LLM design often lacks this integration, leading to systems that may not fully align with human cognitive capabilities and limitations. Insufficient focus on incorporating cognitive science methods exacerbates biases in LLM outputs, while inconsistent application of user-centered design principles results in sub-optimal user experiences. To address these challenges, our position paper explores the critical integration of cognitive ergonomics principles into LLM design, aiming to provide a comprehensive framework and practical guidelines for ethical LLM development. Through our contributions, we seek to advance understanding and practice in integrating cognitive ergonomics into LLM systems, fostering safer, more reliable, and ethically sound human-AI interactions.

Activation sparsity refers to the existence of considerable weakly-contributed elements among activation outputs. As a prevalent property of the models using the ReLU activation function, activation sparsity has been proven a promising paradigm to boost model inference efficiency. Nevertheless, most large language models (LLMs) adopt activation functions without intrinsic activation sparsity (e.g., GELU and Swish). Some recent efforts have explored introducing ReLU or its variants as the substitutive activation function to help LLMs achieve activation sparsity and inference acceleration, but few can simultaneously obtain high sparsity and comparable model performance. This paper introduces a simple and effective sparsification method named "ProSparse" to push LLMs for higher activation sparsity while maintaining comparable performance. Specifically, after substituting the activation function of LLMs with ReLU, ProSparse adopts progressive sparsity regularization with a factor smoothly increasing along the multi-stage sine curves. This can enhance activation sparsity and mitigate performance degradation by avoiding radical shifts in activation distributions. With ProSparse, we obtain high sparsity of 89.32% for LLaMA2-7B, 88.80% for LLaMA2-13B, and 87.89% for end-size MiniCPM-1B, respectively, achieving comparable performance to their original Swish-activated versions. These present the most sparsely activated models among open-source LLaMA versions and competitive end-size models, considerably surpassing ReluLLaMA-7B (66.98%) and ReluLLaMA-13B (71.56%). Our inference acceleration experiments further demonstrate the significant practical acceleration potential of LLMs with higher activation sparsity, obtaining up to 4.52$\times$ inference speedup.

Leveraging Large Language Models' remarkable proficiency in text-based tasks, recent works on Multi-modal LLMs (MLLMs) extend them to other modalities like vision and audio. However, the progress in these directions has been mostly focused on tasks that only require a coarse-grained understanding of the audio-visual semantics. We present Meerkat, an audio-visual LLM equipped with a fine-grained understanding of image and audio both spatially and temporally. With a new modality alignment module based on optimal transport and a cross-attention module that enforces audio-visual consistency, Meerkat can tackle challenging tasks such as audio referred image grounding, image guided audio temporal localization, and audio-visual fact-checking. Moreover, we carefully curate a large dataset AVFIT that comprises 3M instruction tuning samples collected from open-source datasets, and introduce MeerkatBench that unifies five challenging audio-visual tasks. We achieve state-of-the-art performance on all these downstream tasks with a relative improvement of up to 37.12%.

Large language models for code (i.e., code LLMs) have shown strong code understanding and generation capabilities. To evaluate the capabilities of code LLMs in various aspects, many benchmarks have been proposed (e.g., HumanEval and ClassEval). Code reasoning is one of the most essential abilities of code LLMs, but existing benchmarks for code reasoning are not sufficient. Typically, they focus on predicting the input and output of a program, ignoring the evaluation of the intermediate behavior during program execution, as well as the logical consistency (e.g., the model should not give the correct output if the prediction of execution path is wrong) when performing the reasoning. To address these problems, in this paper, we propose a framework, namely REval, for evaluating code reasoning abilities and consistency of code LLMs with program execution. We utilize existing code benchmarks and adapt them to new benchmarks within our framework. A large-scale empirical study is conducted and most LLMs show unsatisfactory performance on both Runtime Behavior Reasoning (i.e., an average accuracy of 44.4%) and Incremental Consistency Evaluation (i.e., an average IC score of 10.3). Evaluation results of current code LLMs reflect the urgent need for the community to strengthen the code reasoning capability of code LLMs. Our code, data, and \newname leaderboard are available at https://r-eval.github.io.

Length generalization refers to the ability to extrapolate from short training sequences to long test sequences and is a challenge for current large language models. While prior work has proposed some architecture or data format changes to achieve length generalization, these proposals typically apply to a limited set of tasks. Building on prior scratchpad and Chain-of-Thought (CoT) techniques, we propose Turing Programs, a novel CoT strategy that decomposes an algorithmic task into steps mimicking the computation of a Turing Machine. This framework is both universal, as it can accommodate any algorithmic task, and simple, requiring only copying text from the context with small modifications. We show that by using Turing Programs, we obtain robust length generalization on a range of algorithmic tasks: addition, multiplication and in-context SGD. We then demonstrate that transformers achieve length generalization on random Turing Programs, suggesting that length generalization is possible for any algorithmic task. Finally, we theoretically prove that transformers can implement Turing Programs, constructing a simple RASP (Weiss et al. [53]) program that simulates an arbitrary Turing machine.

Using language models as a remote service entails sending private information to an untrusted provider. In addition, potential eavesdroppers can intercept the messages, thereby exposing the information. In this work, we explore the prospects of avoiding such data exposure at the level of text manipulation. We focus on text classification models, examining various token mapping and contextualized manipulation functions in order to see whether classifier accuracy may be maintained while keeping the original text unrecoverable. We find that although some token mapping functions are easy and straightforward to implement, they heavily influence performance on the downstream task, and via a sophisticated attacker can be reconstructed. In comparison, the contextualized manipulation provides an improvement in performance.