Optimizing the Key-Value (KV) cache of the Large Language Model (LLM) has been considered critical to saving the cost of inference. Most of the existing KV-cache compression algorithms attempted to sparsify the sequence of tokens by taking advantage of the different importance of tokens. In this work, we found that by identifying the importance of attention layers, we could optimize the KV-cache jointly from two dimensions. Based on our observations regarding layer-wise importance in inference, we propose SqueezeAttention to precisely optimize the allocation of KV-cache budget among layers on-the-fly and then incorporate three representative token sparsification algorithms to compress the KV-cache for each layer with its very own budget. By optimizing the KV-cache from both sequence's and layer's dimensions, SqueezeAttention achieves around 30% to 70% of the memory reductions and up to 2.2 times of throughput improvements in a wide range of LLMs and benchmarks. The code is available at https://github.com/hetailang/SqueezeAttention.

Knowledge graphs contain informative factual knowledge but are considered incomplete. To answer complex queries under incomplete knowledge, learning-based Complex Query Answering (CQA) models are proposed to directly learn from the query-answer samples to avoid the direct traversal of incomplete graph data. Existing works formulate the training of complex query answering models as multi-task learning and require a large number of training samples. In this work, we explore the compositional structure of complex queries and argue that the different logical operator types, rather than the different complex query types, are the key to improving generalizability. Accordingly, we propose a meta-learning algorithm to learn the meta-operators with limited data and adapt them to different instances of operators under various complex queries. Empirical results show that learning meta-operators is more effective than learning original CQA or meta-CQA models.

We explore how iterative revising a chain of thoughts with the help of information retrieval significantly improves large language models' reasoning and generation ability in long-horizon generation tasks, while hugely mitigating hallucination. In particular, the proposed method -- *retrieval-augmented thoughts* (RAT) -- revises each thought step one by one with retrieved information relevant to the task query, the current and the past thought steps, after the initial zero-shot CoT is generated. Applying RAT to GPT-3.5, GPT-4, and CodeLLaMA-7b substantially improves their performances on various long-horizon generation tasks; on average of relatively increasing rating scores by 13.63% on code generation, 16.96% on mathematical reasoning, 19.2% on creative writing, and 42.78% on embodied task planning. The demo page can be found at https://craftjarvis.github.io/RAT

Recent developments have underscored the critical role of \textit{differential privacy} (DP) in safeguarding individual data for training machine learning models. However, integrating DP oftentimes incurs significant model performance degradation due to the perturbation introduced into the training process, presenting a formidable challenge in the {differentially private machine learning} (DPML) field. To this end, several mitigative efforts have been proposed, typically revolving around formulating new DPML algorithms or relaxing DP definitions to harmonize with distinct contexts. In spite of these initiatives, the diminishment induced by DP on models, particularly large-scale models, remains substantial and thus, necessitates an innovative solution that adeptly circumnavigates the consequential impairment of model utility. In response, we introduce DPAdapter, a pioneering technique designed to amplify the model performance of DPML algorithms by enhancing parameter robustness. The fundamental intuition behind this strategy is that models with robust parameters are inherently more resistant to the noise introduced by DP, thereby retaining better performance despite the perturbations. DPAdapter modifies and enhances the sharpness-aware minimization (SAM) technique, utilizing a two-batch strategy to provide a more accurate perturbation estimate and an efficient gradient descent, thereby improving parameter robustness against noise. Notably, DPAdapter can act as a plug-and-play component and be combined with existing DPML algorithms to further improve their performance. Our experiments show that DPAdapter vastly enhances state-of-the-art DPML algorithms, increasing average accuracy from 72.92\% to 77.09\% with a privacy budget of $\epsilon=4$.

The further possibilities in data management (Wang et al., 2022; uncertain nature of knowledge is widely observed Ren et al., 2023). in the real world, but does not align seamlessly with the first-order logic underpinning existing Uncertain knowledge is widely observed from the daily studies. To bridge this gap, we study the setting events (Zhang et al., 2020) to the interaction of biological of soft queries on uncertain knowledge, which systems (Szklarczyk et al., 2023). Besides, uncertainty is is motivated by the establishment of soft constraint also particularly pervasive in KGs because KGs are constructed programming. We further propose an MLbased by information extraction models that could introduce approach with both forward inference and errors (Angeli et al., 2015; Ponte & Croft, 2017) backward calibration to answer soft queries on and from large corpses that could be noisy (Carlson et al., large-scale, incomplete, and uncertain knowledge 2010). To represent the uncertain knowledge, confidence graphs. Theoretical discussions present that our values p are associated with triples in many well-established methods share the same complexity as state-ofthe-art KGs (Carlson et al., 2010; Speer et al., 2017; Szklarczyk inference algorithms for first-order queries.

Recent progress in LLMs discussion suggests that multi-agent discussion improves the reasoning abilities of LLMs. In this work, we reevaluate this claim through systematic experiments, where we propose a novel group discussion framework to enrich the set of discussion mechanisms. Interestingly, our results show that a single-agent LLM with strong prompts can achieve almost the same performance as the best existing discussion approach on a wide range of reasoning tasks and backbone LLMs. We observe that the multi-agent discussion performs better than a single agent only when there is no demonstration in the prompt. Further study reveals the common interaction mechanisms of LLMs during the discussion.

Complex logical query answering is a challenging task in knowledge graphs (KGs) that has been widely studied. The ability to perform complex logical reasoning is essential and supports various graph reasoning-based downstream tasks, such as search engines. Recent approaches are proposed to represent KG entities and logical queries into embedding vectors and find answers to logical queries from the KGs. However, existing proposed methods mainly focus on querying a single KG and cannot be applied to multiple graphs. In addition, directly sharing KGs with sensitive information may incur privacy risks, making it impractical to share and construct an aggregated KG for reasoning to retrieve query answers. Thus, it remains unknown how to answer queries on multi-source KGs. An entity can be involved in various knowledge graphs and reasoning on multiple KGs and answering complex queries on multi-source KGs is important in discovering knowledge cross graphs. Fortunately, federated learning is utilized in knowledge graphs to collaboratively learn representations with privacy preserved. Federated knowledge graph embeddings enrich the relations in knowledge graphs to improve the representation quality. However, these methods only focus on one-hop relations and cannot perform complex reasoning tasks. In this paper, we apply federated learning to complex query-answering tasks to reason over multi-source knowledge graphs while preserving privacy. We propose a Federated Complex Query Answering framework (FedCQA), to reason over multi-source KGs avoiding sensitive raw data transmission to protect privacy. We conduct extensive experiments on three real-world datasets and evaluate retrieval performance on various types of complex queries.

Well-designed prompts have demonstrated the potential to guide text-to-image models in generating amazing images. Although existing prompt engineering methods can provide high-level guidance, it is challenging for novice users to achieve the desired results by manually entering prompts due to a discrepancy between novice-user-input prompts and the model-preferred prompts. To bridge the distribution gap between user input behavior and model training datasets, we first construct a novel Coarse-Fine Granularity Prompts dataset (CFP) and propose a novel User-Friendly Fine-Grained Text Generation framework (UF-FGTG) for automated prompt optimization. For CFP, we construct a novel dataset for text-to-image tasks that combines coarse and fine-grained prompts to facilitate the development of automated prompt generation methods. For UF-FGTG, we propose a novel framework that automatically translates user-input prompts into model-preferred prompts. Specifically, we propose a prompt refiner that continually rewrites prompts to empower users to select results that align with their unique needs. Meanwhile, we integrate image-related loss functions from the text-to-image model into the training process of text generation to generate model-preferred prompts. Additionally, we propose an adaptive feature extraction module to ensure diversity in the generated results. Experiments demonstrate that our approach is capable of generating more visually appealing and diverse images than previous state-of-the-art methods, achieving an average improvement of 5% across six quality and aesthetic metrics.

The rapidly evolving multimodal Large Language Models (LLMs) urgently require new benchmarks to uniformly evaluate their performance on understanding and textually describing music. However, due to semantic gaps between Music Information Retrieval (MIR) algorithms and human understanding, discrepancies between professionals and the public, and low precision of annotations, existing music description datasets cannot serve as benchmarks. To this end, we present MuChin, the first open-source music description benchmark in Chinese colloquial language, designed to evaluate the performance of multimodal LLMs in understanding and describing music. We established the Caichong Music Annotation Platform (CaiMAP) that employs an innovative multi-person, multi-stage assurance method, and recruited both amateurs and professionals to ensure the precision of annotations and alignment with popular semantics. Utilizing this method, we built a dataset with multi-dimensional, high-precision music annotations, the Caichong Music Dataset (CaiMD), and carefully selected 1,000 high-quality entries to serve as the test set for MuChin. Based on MuChin, we analyzed the discrepancies between professionals and amateurs in terms of music description, and empirically demonstrated the effectiveness of annotated data for fine-tuning LLMs. Ultimately, we employed MuChin to evaluate existing music understanding models on their ability to provide colloquial descriptions of music. All data related to the benchmark and the code for scoring have been open-sourced.

The ever-growing ecosystem of LLMs has posed a challenge in selecting the most appropriate pre-trained model to fine-tune amidst a sea of options. Given constrained resources, fine-tuning all models and making selections afterward is unrealistic. In this work, we formulate this resource-constrained selection task into predicting fine-tuning performance and illustrate its natural connection with scaling laws. Unlike pre-training, We find that the fine-tuning scaling curve includes not just the well-known "power phase" but also the previously unobserved "pre-power phase". We also explain why existing scaling laws fail to capture this phase transition phenomenon both theoretically and empirically. To address this, we introduce the concept of "pre-learned data size" into our rectified scaling law, which overcomes theoretical limitations and fits experimental results much better. By leveraging our law, we propose a novel LLM selection algorithm that selects the near-optimal model with hundreds of times less resource consumption, while other methods may provide negatively correlated selection.