In recent years, Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide array of text-centric tasks. However, their `large' scale introduces significant computational and storage challenges, particularly in managing the key-value states of the transformer, which limits their wider applicability. Therefore, we propose to adaptively release resources from caches and rebuild the necessary key-value states. Particularly, we accomplish this by a lightweight controller module to approximate an ideal top-$K$ sparse attention. This module retains the tokens with the highest top-$K$ attention weights and simultaneously rebuilds the discarded but necessary tokens, which may become essential for future decoding. Comprehensive experiments in natural language generation and modeling reveal that our method is not only competitive with full attention in terms of performance but also achieves a significant throughput improvement of up to 221.8%. The code for replication is available on the https://github.com/WHUIR/ADORE.

Clinical documentation is an important aspect of clinicians' daily work and often demands a significant amount of time. The BioNLP 2024 Shared Task on Streamlining Discharge Documentation (Discharge Me!) aims to alleviate this documentation burden by automatically generating discharge summary sections, including brief hospital course and discharge instruction, which are often time-consuming to synthesize and write manually. We approach the generation task by fine-tuning multiple open-sourced language models (LMs), including both decoder-only and encoder-decoder LMs, with various configurations on input context. We also examine different setups for decoding algorithms, model ensembling or merging, and model specialization. Our results show that conditioning on the content of discharge summary prior to the target sections is effective for the generation task. Furthermore, we find that smaller encoder-decoder LMs can work as well or even slightly better than larger decoder based LMs fine-tuned through LoRA. The model checkpoints from our team (aehrc) are openly available.

Large language models (LLMs) typically utilize the top-k contexts from a retriever in retrieval-augmented generation (RAG). In this work, we propose a novel instruction fine-tuning framework RankRAG, which instruction-tunes a single LLM for the dual purpose of context ranking and answer generation in RAG. In particular, the instruction-tuned LLMs work surprisingly well by adding a small fraction of ranking data into the training blend, and outperform existing expert ranking models, including the same LLM exclusively fine-tuned on a large amount of ranking data. For generation, we compare our model with many strong baselines, including GPT-4-0613, GPT-4-turbo-2024-0409, and ChatQA-1.5, an open-sourced model with the state-of-the-art performance on RAG benchmarks. Specifically, our Llama3-RankRAG significantly outperforms Llama3-ChatQA-1.5 and GPT-4 models on nine knowledge-intensive benchmarks. In addition, it also performs comparably to GPT-4 on five RAG benchmarks in the biomedical domain without instruction fine-tuning on biomedical data, demonstrating its superb capability for generalization to new domains.

Recently, many studies have demonstrated that exclusively incorporating OCR-derived text and spatial layouts with large language models (LLMs) can be highly effective for document understanding tasks. However, existing methods that integrate spatial layouts with text have limitations, such as producing overly long text sequences or failing to fully leverage the autoregressive traits of LLMs. In this work, we introduce Interleaving Layout and Text in a Large Language Model (LayTextLLM)} for document understanding. In particular, LayTextLLM projects each bounding box to a single embedding and interleaves it with text, efficiently avoiding long sequence issues while leveraging autoregressive traits of LLMs. LayTextLLM not only streamlines the interaction of layout and textual data but also shows enhanced performance in Key Information Extraction (KIE) and Visual Question Answering (VQA). Comprehensive benchmark evaluations reveal significant improvements, with a 27.0% increase on KIE tasks and 24.1% on VQA tasks compared to previous state-of-the-art document understanding MLLMs, as well as a 15.5% improvement over other SOTA OCR-based LLMs on KIE tasks.

As Large Language Models (LLMs) are increasingly deployed to handle various natural language processing (NLP) tasks, concerns regarding the potential negative societal impacts of LLM-generated content have also arisen. To evaluate the biases exhibited by LLMs, researchers have recently proposed a variety of datasets. However, existing bias evaluation efforts often focus on only a particular type of bias and employ inconsistent evaluation metrics, leading to difficulties in comparison across different datasets and LLMs. To address these limitations, we collect a variety of datasets designed for the bias evaluation of LLMs, and further propose CEB, a Compositional Evaluation Benchmark with 11,004 samples that cover different types of bias across different social groups and tasks. The curation of CEB is based on our newly proposed compositional taxonomy, which characterizes each dataset from three dimensions: bias types, social groups, and tasks. By combining the three dimensions, we develop a comprehensive evaluation strategy for the bias in LLMs. Our experiments demonstrate that the levels of bias vary across these dimensions, thereby providing guidance for the development of specific bias mitigation methods.

The use of an efficient coverage planning method is key for autonomous navigation in agricultural environments, where a robot must cover large areas of crops. This paper generally reviews the current state of the art of coverage path planning methods. Two widely used techniques applicable to agricultural environments are described in detail. The first consists of breaking down a complex field with obstacles into simpler subregions known as cells, to subsequently generate a coverage pattern in each of them. The second analyzes spaces composed of parallel strips through which the robot must circulate, in order to find the optimal order of visiting strips that minimizes the total distance traveled. Additionally, the combination of both techniques is discussed in order to obtain a more efficient global coverage plan. This analysis was conceived to be implemented with the soybean crop weeding robot developed at CIFASIS (CONICET-UNR).

Knowledge Graph Embedding (KGE) is a common method for Knowledge Graphs (KGs) to serve various artificial intelligence tasks. The suitable dimensions of the embeddings depend on the storage and computing conditions of the specific application scenarios. Once a new dimension is required, a new KGE model needs to be trained from scratch, which greatly increases the training cost and limits the efficiency and flexibility of KGE in serving various scenarios. In this work, we propose a novel KGE training framework MED, through which we could train once to get a croppable KGE model applicable to multiple scenarios with different dimensional requirements, sub-models of the required dimensions can be cropped out of it and used directly without any additional training. In MED, we propose a mutual learning mechanism to improve the low-dimensional sub-models performance and make the high-dimensional sub-models retain the capacity that low-dimensional sub-models have, an evolutionary improvement mechanism to promote the high-dimensional sub-models to master the knowledge that the low-dimensional sub-models can not learn, and a dynamic loss weight to balance the multiple losses adaptively. Experiments on 3 KGE models over 4 standard KG completion datasets, 3 real application scenarios over a real-world large-scale KG, and the experiments of extending MED to the language model BERT show the effectiveness, high efficiency, and flexible extensibility of MED.

This paper introduces Neurocache, an approach to extend the effective context size of large language models (LLMs) using an external vector cache to store its past states. Like recent vector retrieval approaches, Neurocache uses an efficient k-nearest-neighbor (kNN) algorithm to retrieve relevant past states and incorporate them into the attention process. Neurocache improves upon previous methods by (1) storing compressed states, which reduces cache size; (2) performing a single retrieval operation per token which increases inference speed; and (3) extending the retrieval window to neighboring states, which improves both language modeling and downstream task accuracy. Our experiments show the effectiveness of Neurocache both for models trained from scratch and for pre-trained models such as Llama2-Figure 1: Performance and Scalability of Neurocache 7B and Mistral-7B when enhanced with the vs. Memorizing Transformers (Wu et al., 2022) on cache mechanism. We also compare Neurocache PG-19: The graph illustrates Neurocache's consistently with text retrieval methods and show lower token perplexity and faster inference times across improvements in single-document questionanswering various cache sizes on the Project Gutenberg-19 dataset, and few-shot learning tasks.

The computational challenges of Large Language Model (LLM) inference remain a significant barrier to their widespread deployment, especially as prompt lengths continue to increase. Due to the quadratic complexity of the attention computation, it takes 30 minutes for an 8B LLM to process a prompt of 1M tokens (i.e., the pre-filling stage) on a single A100 GPU. Existing methods for speeding up prefilling often fail to maintain acceptable accuracy or efficiency when applied to long-context LLMs. To address this gap, we introduce MInference (Milliontokens Inference), a sparse calculation method designed to accelerate pre-filling of long-sequence processing. Specifically, we identify three unique patterns in long-context attention matrices-the A-shape, Vertical-Slash, and Block-Sparsethat can be leveraged for efficient sparse computation on GPUs. We determine the optimal pattern for each attention head offline and dynamically build sparse indices based on the assigned pattern during inference. With the pattern and sparse indices, we perform efficient sparse attention calculations via our optimized GPU kernels to significantly reduce the latency in the pre-filling stage of long-context LLMs. Our proposed technique can be directly applied to existing LLMs without any modifications to the pre-training setup or additional fine-tuning. By evaluating on a wide range of downstream tasks, including InfiniteBench, RULER, PG-19, and Needle In A Haystack, and models including LLaMA-3-1M, GLM4-1M, Yi-200K, Phi-3-128K, and Qwen2-128K, we demonstrate that MInference effectively reduces inference latency by up to 10x for pre-filling on an A100, while maintaining accuracy. Our code is available at https://aka.ms/MInference.

Software systems often have numerous configuration options that can be adjusted to meet different performance requirements. However, understanding the combined impact of these options on performance is often challenging, especially with limited real-world data. To tackle this issue, deep learning techniques have gained popularity due to their ability to capture complex relationships even with limited samples. This thesis begins with a systematic literature review of deep learning techniques in configuration performance modeling, analyzing 85 primary papers out of 948 searched papers. It identifies knowledge gaps and sets three objectives for the thesis. The first knowledge gap is the lack of understanding about which encoding scheme is better and in what circumstances. To address this, the thesis conducts an empirical study comparing three popular encoding schemes. Actionable suggestions are provided to support more reliable decisions. Another knowledge gap is the sparsity inherited from the configuration landscape. To handle this, the thesis proposes a model-agnostic and sparsity-robust framework called DaL, which uses a "divide-and-learn" approach. DaL outperforms state-of-the-art approaches in accuracy improvement across various real-world systems. The thesis also addresses the limitation of predicting under static environments by proposing a sequential meta-learning framework called SeMPL. Unlike traditional meta-learning frameworks, SeMPL trains meta-environments in a specialized order, resulting in significantly improved prediction accuracy in multi-environment scenarios. Overall, the thesis identifies and addresses critical knowledge gaps in deep performance learning, significantly advancing the accuracy of performance prediction.