Large Language Models (LLMs) have achieved remarkable success and have been applied across various scientific fields, including chemistry. However, many chemical tasks require the processing of visual information, which cannot be successfully handled by existing chemical LLMs. This brings a growing need for models capable of integrating multimodal information in the chemical domain. In this paper, we introduce \textbf{ChemVLM}, an open-source chemical multimodal large language model specifically designed for chemical applications. ChemVLM is trained on a carefully curated bilingual multimodal dataset that enhances its ability to understand both textual and visual chemical information, including molecular structures, reactions, and chemistry examination questions. We develop three datasets for comprehensive evaluation, tailored to Chemical Optical Character Recognition (OCR), Multimodal Chemical Reasoning (MMCR), and Multimodal Molecule Understanding tasks. We benchmark ChemVLM against a range of open-source and proprietary multimodal large language models on various tasks. Experimental results demonstrate that ChemVLM achieves competitive performance across all evaluated tasks. Our model can be found at https://huggingface.co/AI4Chem/ChemVLM-26B.

Recent advancements in retrieval-augmented generation (RAG) have demonstrated impressive performance in the question-answering (QA) task. However, most previous works predominantly focus on text-based answers. While some studies address multimodal data, they still fall short in generating comprehensive multimodal answers, particularly for explaining concepts or providing step-by-step tutorials on how to accomplish specific goals. This capability is especially valuable for applications such as enterprise chatbots and settings such as customer service and educational systems, where the answers are sourced from multimodal data. In this paper, we introduce a simple and effective framework named MuRAR (Multimodal Retrieval and Answer Refinement). MuRAR enhances text-based answers by retrieving relevant multimodal data and refining the responses to create coherent multimodal answers. This framework can be easily extended to support multimodal answers in enterprise chatbots with minimal modifications. Human evaluation results indicate that multimodal answers generated by MuRAR are more useful and readable compared to plain text answers.

Audio generation has achieved remarkable progress with the advance of sophisticated generative models, such as diffusion models (DMs) and autoregressive (AR) models. However, due to the naturally significant sequence length of audio, the efficiency of audio generation remains an essential issue to be addressed, especially for AR models that are incorporated in large language models (LLMs). In this paper, we analyze the token length of audio tokenization and propose a novel \textbf{S}cale-level \textbf{A}udio \textbf{T}okenizer (SAT), with improved residual quantization. Based on SAT, a scale-level \textbf{A}coustic \textbf{A}uto\textbf{R}egressive (AAR) modeling framework is further proposed, which shifts the next-token AR prediction to next-scale AR prediction, significantly reducing the training cost and inference time. To validate the effectiveness of the proposed approach, we comprehensively analyze design choices and demonstrate the proposed AAR framework achieves a remarkable \textbf{35}$\times$ faster inference speed and +\textbf{1.33} Fr\'echet Audio Distance (FAD) against baselines on the AudioSet benchmark. Code: \url{https://github.com/qiuk2/AAR}.

In this article, we explore the challenges and evolution of two key technologies in the current field of AI: Vision Transformer model and Large Language Model (LLM). Vision Transformer captures global information by splitting images into small pieces and leveraging Transformer's multi-head attention mechanism, but its high reference count and compute overhead limit deployment on mobile devices. At the same time, the rapid development of LLM has revolutionized natural language processing, but it also faces huge deployment challenges. To address these issues, we investigate model pruning techniques, with a particular focus on how to reduce redundant parameters without losing accuracy to accommodate personalized data and resource-constrained environments. In this paper, a new layered pruning strategy is proposed to distinguish the personalized layer from the common layer by compressed sensing and random sampling, thus significantly reducing the model parameters. Our experimental results show that the introduced step buffering mechanism further improves the accuracy of the model after pruning, providing new directions and possibilities for the deployment of efficient and personalized AI models on mobile devices in the future.

Accurate traffic forecasting is crucial for effective urban planning and transportation management, enabling efficient resource allocation and enhanced travel experiences. However, existing models often face limitations in generalization, struggling with zero-shot prediction on unseen regions and cities, as well as diminished long-term accuracy. This is primarily due to the inherent challenges in handling the spatial and temporal heterogeneity of traffic data, coupled with the significant distribution shift across time and space. In this work, we aim to unlock new possibilities for building versatile, resilient and adaptive spatio-temporal foundation models for traffic prediction. To achieve this goal, we introduce a novel foundation model, named OpenCity, that can effectively capture and normalize the underlying spatio-temporal patterns from diverse data characteristics, facilitating zero-shot generalization across diverse urban environments. OpenCity integrates the Transformer architecture with graph neural networks to model the complex spatio-temporal dependencies in traffic data. By pre-training OpenCity on large-scale, heterogeneous traffic datasets, we enable the model to learn rich, generalizable representations that can be seamlessly applied to a wide range of traffic forecasting scenarios. Experimental results demonstrate that OpenCity exhibits exceptional zero-shot predictive performance. Moreover, OpenCity showcases promising scaling laws, suggesting the potential for developing a truly one-for-all traffic prediction solution that can adapt to new urban contexts with minimal overhead. We made our proposed OpenCity model open-source and it is available at the following link: https://github.com/HKUDS/OpenCity.

Multilingual large language models (MLLMs) struggle to answer questions posed in non-dominant languages, even though they have acquired the relevant knowledge from their dominant language corpus. In contrast, human multilinguals can overcome such non-native language context limitations through Positive Native Language Transfer (PNLT). Inspired by the process of PNLT, we analogize the dominant language of MLLMs to the native language of human multilinguals, and propose Native Language Prompting (NatLan) to simulate the PNLT observed in human multilinguals. It explicitly creates native language contexts for MLLMs to facilitate the elicitation of the rich native language knowledge during question-answering, unlocking the limitations imposed by non-native language contexts. By employing multi-MLLM collaboration, NatLan reduces the workload on each MLLM in simulating PNLT and refines semantic transfer. On the C-Eval benchmark, NatLan provides up to a 10.1% average accuracy improvement and up to a 5.0% increase in the hard-level subset across five MLLMs, surpassing all top-notch related methods. Our code is available at https://github.com/AnonyNLP/NatLan.

The Mixture of Experts (MoE) framework has become a popular architecture for large language models due to its superior performance over dense models. However, training MoEs from scratch in a large-scale regime is prohibitively expensive. Existing methods mitigate this by pre-training multiple dense expert models independently and using them to initialize an MoE. This is done by using experts' feed-forward network (FFN) to initialize the MoE's experts while merging other parameters. However, this method limits the reuse of dense model parameters to only the FFN layers, thereby constraining the advantages when "upcycling" these models into MoEs. We propose BAM (Branch-Attend-Mix), a simple yet effective method that addresses this shortcoming. BAM makes full use of specialized dense models by not only using their FFN to initialize the MoE layers but also leveraging experts' attention parameters fully by initializing them into a soft-variant of Mixture of Attention (MoA) layers. We explore two methods for upcycling attention parameters: 1) initializing separate attention experts from dense models including all attention parameters for the best model performance; and 2) sharing key and value parameters across all experts to facilitate for better inference efficiency. To further improve efficiency, we adopt a parallel attention transformer architecture to MoEs, which allows the attention experts and FFN experts to be computed concurrently. Our experiments on seed models ranging from 590 million to 2 billion parameters demonstrate that BAM surpasses baselines in both perplexity and downstream task performance, within the same computational and data constraints.

The rapid growth in the parameters of large language models (LLMs) has made inference latency a fundamental bottleneck, limiting broader application of LLMs. Speculative decoding represents a lossless approach to accelerate inference through a guess-and-verify paradigm, leveraging the parallel capabilities of modern hardware. Some speculative decoding methods rely on additional structures to guess draft tokens, such as small models or parameter-efficient architectures, which need extra training before use. Alternatively, retrieval-based train-free techniques build libraries from pre-existing corpora or by n-gram generation. However, they face challenges like large storage requirements, time-consuming retrieval, and limited adaptability. Observing that candidate tokens generated during the decoding process are likely to reoccur in future sequences, we propose Token Recycling. This approach stores candidate tokens in an adjacency matrix and employs a breadth-first search (BFS)-like algorithm on the matrix to construct a draft tree. The tree is then validated through tree attention. New candidate tokens from the decoding process are then used to update the matrix. Token Recycling requires \textless2MB of additional storage and achieves approximately 2x speedup across all sizes of LLMs. It significantly outperforms existing train-free methods by 30\% and even a training method by 25\%. It can be directly applied to any existing LLMs and tasks without the need for adaptation.

The pre-training cost of large language models (LLMs) is prohibitive. One cutting-edge approach to reduce the cost is zero-shot weight transfer, also known as model growth for some cases, which magically transfers the weights trained in a small model to a large model. However, there are still some theoretical mysteries behind the weight transfer. In this paper, inspired by prior applications of mean field theory to neural network dynamics, we introduce a mean field ansatz to provide a theoretical explanation for weight transfer. Specifically, we propose the row-column (RC) ansatz under the mean field point of view, which describes the measure structure of the weights in the neural network (NN) and admits a close measure dynamic. Thus, the weights of different sizes NN admit a common distribution under proper assumptions, and weight transfer methods can be viewed as sampling methods. We empirically validate the RC ansatz by exploring simple MLP examples and LLMs such as GPT-3 and Llama-3.1. We show the mean-field point of view is adequate under suitable assumptions which can provide theoretical support for zero-shot weight transfer.

Clinical question answering systems have the potential to provide clinicians with relevant and timely answers to their questions. Nonetheless, despite the advances that have been made, adoption of these systems in clinical settings has been slow. One issue is a lack of question-answering datasets which reflect the real-world needs of health professionals. In this work, we present RealMedQA, a dataset of realistic clinical questions generated by humans and an LLM. We describe the process for generating and verifying the QA pairs and assess several QA models on BioASQ and RealMedQA to assess the relative difficulty of matching answers to questions. We show that the LLM is more cost-efficient for generating "ideal" QA pairs. Additionally, we achieve a lower lexical similarity between questions and answers than BioASQ which provides an additional challenge to the top two QA models, as per the results. Introduction Clinical question answering (QA) systems could allow clinicians to find timely and relevant answers to questions occurring during consultations in real-time [1, 2, 3, 4, 5].