The efficiency of large vision-language models (LVLMs) is constrained by the computational bottleneck of the attention mechanism during the prefill phase and the memory bottleneck of fetching the key-value (KV) cache in the decoding phase, particularly in scenarios involving high-resolution images or videos. Visual content often exhibits substantial redundancy, resulting in highly sparse attention maps within LVLMs. This sparsity can be leveraged to accelerate attention computation or compress the KV cache through various approaches. However, most studies focus on addressing only one of these bottlenecks and do not adequately support dynamic adjustment of sparsity concerning distinct layers or tasks. In this paper, we present ZipVL, an efficient inference framework designed for LVLMs through a dynamic ratio allocation strategy of important tokens. This ratio is adaptively determined based on the layer-specific distribution of attention scores, rather than fixed hyper-parameters, thereby improving efficiency for less complex tasks while maintaining high performance for more challenging ones. Then we select important tokens based on their normalized attention scores and perform sparse attention mechanism solely on those important tokens, reducing the latency in the prefill phase. Tokens deemed less important will be discarded to reduce KV cache size, alleviating the memory bottleneck in the decoding phase. Our experiments demonstrate that ZipVL can accelerate the prefill phase by 2.3$\times$ and improve decoding throughput by 2.8$\times$, with a minimal accuracy reduction of only 0.5\% on VQAv2 benchmark over LLaVA-Next-13B model, effectively enhancing the generation efficiency of LVLMs.

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. The code is available at https://github.com/Infini-AI-Lab/MagicPIG.

To uniformly predict 1D and 2D time series, we generalize next token prediction, predominantly adopted for causal generation of 1D sequences, to multivariate next token prediction. The proposed paradigm uniformly formulates various forecasting scenarios as a long-context generation problem. We opt for the generative Transformer, which can capture global-range and causal dependencies while providing contextual flexibility, to implement unified forecasting on univariate series characterized by non-stationarity, multivariate time series with complicated dynamics and correlations, and covariate-informed contexts that include both endogenous and exogenous time series. Technically, we propose a universal TimeAttention to facilitate generative Transformers on multiple time series, which can effectively capture fine-grained intra-and inter-series dependencies of flattened time series tokens (patches), and is further enhanced by deftly designed position embeddings for the temporal and variable dimensions. Timer-XL achieves state-of-the-art performance across challenging forecasting benchmarks through a unified approach. Based on large-scale pre-training, Timer-XL also demonstrates notable zero-shot performance, making it a promising architecture for large time series models. Transformers have contributed significantly to the fields of natural language and computer vision (Radford et al., 2018; Dosovitskiy et al., 2020), and been extensively applied in time series forecasting, becoming the foundation of specialized forecasters (Zhou et al., 2021; Wu et al., 2021) and large models (Das et al., 2023). As a typical generative task, the quality of predictions heavily relies on the context (Dai et al., 2019). Reliable predictions are made by thoroughly considering endogenous temporal variations and retrieving relevant exogenous correlations into the context (Box, 2013). Further, the pre-training context length, which serves as an indicator of scaling (Kaplan et al., 2020), determines the maximum input and output of generative Transformers, ultimately enabling long-sequence, high-resolution, and high-frequency generation (Yin et al., 2023; Wang et al., 2024a). However, existing Transformers in the time series field crucially encounter the context bottleneck. As shown in Figure 1, unlike Transformers for natural language and vision that learn dependencies among thousands to millions of tokens (Kirillov et al., 2023; OpenAI, 2023), time-series Transformers typically work around limited contexts of up to hundreds of time series tokens (patches) (Nie et al., 2022). For univariate time series, a short context length leads to an insufficient perception of global tendencies, overlooking widespread non-stationarity in real-world time series (Hyndman, 2018). The excessive reliance on stationarization, such as normalization (Kim et al., 2021), restricts the model capacity and leads to overfitting of Transformers (Liu et al., 2022b).

Multilingual large language models (LLMs) today may not necessarily provide culturally appropriate and relevant responses to its Filipino users. We introduce Kalahi, a cultural LLM evaluation suite collaboratively created by native Filipino speakers. It is composed of 150 high-quality, handcrafted and nuanced prompts that test LLMs for generations that are relevant to shared Filipino cultural knowledge and values. Strong LLM performance in Kalahi indicates a model's ability to generate responses similar to what an average Filipino would say or do in a given situation. We conducted experiments on LLMs with multilingual and Filipino language support. Results show that Kalahi, while trivial for Filipinos, is challenging for LLMs, with the best model answering only 46.0% of the questions correctly compared to native Filipino performance of 89.10%. Thus, Kalahi can be used to accurately and reliably evaluate Filipino cultural representation in LLMs.

Large language models (LLMs), trained on diverse data effectively acquire a breadth of information across various domains. However, their computational complexity, cost, and lack of transparency hinder their direct application for specialised tasks. In fields such as clinical research, acquiring expert annotations or prior knowledge about predictive models is often costly and time-consuming. This study proposes the use of LLMs to elicit expert prior distributions for predictive models. This approach also provides an alternative to in-context learning, where language models are tasked with making predictions directly. In this work, we compare LLM-elicited and uninformative priors, evaluate whether LLMs truthfully generate parameter distributions, and propose a model selection strategy for in-context learning and prior elicitation. Our findings show that LLM-elicited prior parameter distributions significantly reduce predictive error compared to uninformative priors in low-data settings. Applied to clinical problems, this translates to fewer required biological samples, lowering cost and resources. Prior elicitation also consistently outperforms and proves more reliable than in-context learning at a lower cost, making it a preferred alternative in our setting. We demonstrate the utility of this method across various use cases, including clinical applications. For infection prediction, using LLM-elicited priors reduced the number of required labels to achieve the same accuracy as an uninformative prior by 55%, 200 days earlier in the study.

This Large language models (LLMs) have shown remarkable approach is orthogonal to other optimization techniques, capabilities across a wide range of tasks, and could be applicable as LLMs continue from natural language understanding to creative to grow in size and capabilities. We also propose content generation. However, the computational an algorithm to check and define the applicability cost of inference and the associated energy consumption of this technique in different domains, selecting present significant challenges. As the demand the most proper function words set, and analyzing for AI applications continues to grow, these the lose in performance as the percentage of costs are expected to escalate, raising concerns saved tokens increases. Additionally, we provide about sustainability and accessibility (Wu et al., an experimental evaluation in the context of general 2022).

Recent advances in audio understanding tasks leverage the reasoning capabilities of LLMs. However, adapting LLMs to learn audio concepts requires massive training data and substantial computational resources. To address these challenges, Retrieval-Augmented Generation (RAG) retrieves audio-text pairs from a knowledge base (KB) and augments them with query audio to generate accurate textual responses. In RAG, the relevance of the retrieved information plays a crucial role in effectively processing the input. In this paper, we analyze how different retrieval methods and knowledge bases impact the relevance of audio-text pairs and the performance of audio captioning with RAG. We propose generative pair-to-pair retrieval, which uses the generated caption as a text query to accurately find relevant audio-text pairs to the query audio, thereby improving the relevance and accuracy of retrieved information. Additionally, we refine the large-scale knowledge base to retain only audio-text pairs that align with the contextualized intents. Our approach achieves state-of-the-art results on benchmarks including AudioCaps, Clotho, and Auto-ACD, with detailed ablation studies validating the effectiveness of our retrieval and KB construction methods.

Recent advancements in large language models (LLMs) have significantly enhanced their ability to understand both natural language and code, driving their use in tasks like natural language-to-code (NL2Code) and code summarization. However, LLMs are prone to hallucination-outputs that stray from intended meanings. Detecting hallucinations in code summarization is especially difficult due to the complex interplay between programming and natural languages. We introduce a first-of-its-kind dataset with $\sim$10K samples, curated specifically for hallucination detection in code summarization. We further propose a novel Entity Tracing Framework (ETF) that a) utilizes static program analysis to identify code entities from the program and b) uses LLMs to map and verify these entities and their intents within generated code summaries. Our experimental analysis demonstrates the effectiveness of the framework, leading to a 0.73 F1 score. This approach provides an interpretable method for detecting hallucinations by grounding entities, allowing us to evaluate summary accuracy.

The HeartBert model is introduced with three primary objectives: reducing the need for labeled data, minimizing computational resources, and simultaneously improving performance in machine learning systems that analyze Electrocardiogram (ECG) signals. Inspired by Bidirectional Encoder Representations from Transformers (BERT) in natural language processing and enhanced with a self-supervised learning approach, the HeartBert model-built on the RoBERTa architecture-generates sophisticated embeddings tailored for ECG-based projects in the medical domain. To demonstrate the versatility, generalizability, and efficiency of the proposed model, two key downstream tasks have been selected: sleep stage detection and heartbeat classification. HeartBERT-based systems, utilizing bidirectional LSTM heads, are designed to address complex challenges. A series of practical experiments have been conducted to demonstrate the superiority and advancements of HeartBERT, particularly in terms of its ability to perform well with smaller training datasets, reduced learning parameters, and effective performance compared to rival models. The code and data are publicly available at https://github.com/ecgResearch/HeartBert.

While Large Language Models (LLMs) perform exceptionally well in generating accurate outputs, their "black box" nature poses significant challenges to transparency and trust. To address this, the paper proposes the TranspNet pipeline, which integrates symbolic AI with LLMs. By leveraging domain expert knowledge, retrieval-augmented generation (RAG), and formal reasoning frameworks like Answer Set Programming (ASP), TranspNet enhances LLM outputs with structured reasoning and verification.This approach strives to help AI systems deliver results that are as accurate, explainable, and trustworthy as possible, aligning with regulatory expectations for transparency and accountability. TranspNet provides a solution for developing AI systems that are reliable and interpretable, making it suitable for real-world applications where trust is critical.