Compressing KV Cache for Long-Context LLM Inference with Inter-Layer Attention Similarity

The increasing context window size in Large Language Models (LLMs), such as the GPT and LLaMA series, has improved their ability to tackle complex, long-text tasks, but at the cost of inference efficiency, particularly regarding memory and computational complexity. Existing methods, including selective token retention and window-based attention, improve efficiency but risk discarding important tokens needed for future text generation. In this paper, we propose an approach that enhances LLM efficiency without token loss by reducing the memory and computational load of less important tokens, rather than discarding them.We address two challenges: 1) investigating the distribution of important tokens in the context, discovering recent tokens are more important than distant tokens in context, and 2) optimizing resources for distant tokens by sharing attention scores across layers. The experiments show that our method saves $35\%$ KV cache without compromising the performance.

Track Any Peppers: Weakly Supervised Sweet Pepper Tracking Using VLMs

In the Detection and Multi-Object Tracking of Sweet Peppers Challenge, we present Track Any Peppers (TAP) - a weakly supervised ensemble technique for sweet peppers tracking. TAP leverages the zero-shot detection capabilities of vision-language foundation models like Grounding DINO to automatically generate pseudo-labels for sweet peppers in video sequences with minimal human intervention. These pseudo-labels, refined when necessary, are used to train a YOLOv8 segmentation network. To enhance detection accuracy under challenging conditions, we incorporate pre-processing techniques such as relighting adjustments and apply depth-based filtering during post-inference. For object tracking, we integrate the Matching by Segment Anything (MASA) adapter with the BoT-SORT algorithm. Our approach achieves a HOTA score of 80.4%, MOTA of 66.1%, Recall of 74.0%, and Precision of 90.7%, demonstrating effective tracking of sweet peppers without extensive manual effort. This work highlights the potential of foundation models for efficient and accurate object detection and tracking in agricultural settings.

What Factors Affect Multi-Modal In-Context Learning? An In-Depth Exploration

Recently, rapid advancements in Multi-Modal In-Context Learning (MM-ICL) have achieved notable success, which is capable of achieving superior performance across various tasks without requiring additional parameter tuning. However, the underlying rules for the effectiveness of MM-ICL remain under-explored. To fill this gap, this work aims to investigate the research question: "What factors affect the performance of MM-ICL?'' To this end, we investigate extensive experiments on the three core steps of MM-ICL including demonstration retrieval, demonstration ordering, and prompt construction using 6 vision large language models and 20 strategies. Our findings highlight (1) the necessity of a multi-modal retriever for demonstration retrieval, (2) the importance of intra-demonstration ordering over inter-demonstration ordering, and (3) the enhancement of task comprehension through introductory instructions in prompts. We hope this study can serve as a foundational guide for optimizing MM-ICL strategies in future research.

KVSharer: Efficient Inference via Layer-Wise Dissimilar KV Cache Sharing

The development of large language models (LLMs) has significantly expanded model sizes, resulting in substantial GPU memory requirements during inference. The key and value storage of the attention map in the KV (key-value) cache accounts for more than 80\% of this memory consumption. Nowadays, most existing KV cache compression methods focus on intra-layer compression within a single Transformer layer but few works consider layer-wise compression. In this paper, we propose a plug-and-play method called \textit{KVSharer}, which shares the KV cache between layers to achieve layer-wise compression. Rather than intuitively sharing based on higher similarity, we discover a counterintuitive phenomenon: sharing dissimilar KV caches better preserves the model performance. Experiments show that \textit{KVSharer} can reduce KV cache computation by 30\%, thereby lowering memory consumption without significantly impacting model performance and it can also achieve at least 1.3 times generation acceleration. Additionally, we verify that \textit{KVSharer} is compatible with existing intra-layer KV cache compression methods, and combining both can further save memory.

Codebook Design and Performance Analysis for Wideband Beamforming in Terahertz Communications

The codebook-based analog beamforming is appealing for future terahertz (THz) communications since it can generate high-gain directional beams with low-cost phase shifters via low-complexity beam training. However, conventional beamforming codebook design based on array response vectors for narrowband communications may suffer from severe performance loss in wideband systems due to the ``beam squint" effect over frequency. To tackle this issue, we propose in this paper a new codebook design method for analog beamforming in wideband THz systems. In particular, to characterize the analog beamforming performance in wideband systems, we propose a new metric termed wideband beam gain, which is given by the minimum beamforming gain over the entire frequency band given a target angle. Based on this metric, a wideband analog beamforming codebook design problem is formulated for optimally balancing the beamforming gains in both the spatial and frequency domains, and the performance loss of conventional narrowband beamforming in wideband systems is analyzed. To solve the new wideband beamforming codebook design problem, we divide the spatial domain into orthogonal angular zones each associated with one beam, thereby decoupling the codebook design into a zone division sub-problem and a set of beamforming optimization sub-problems each for one zone. For the zone division sub-problem, we propose a bisection method to obtain the optimal boundaries for separating adjacent zones. While for each of the per-zone-based beamforming optimization sub-problems, we further propose an efficient augmented Lagrange method (ALM) to solve it. Numerical results demonstrate the performance superiority of our proposed codebook design for wideband analog beamforming to the narrowband beamforming codebook and also validate our performance analysis.

Evaluating Software Development Agents: Patch Patterns, Code Quality, and Issue Complexity in Real-World GitHub Scenarios

In recent years, AI-based software engineering has progressed from pre-trained models to advanced agentic workflows, with Software Development Agents representing the next major leap. These agents, capable of reasoning, planning, and interacting with external environments, offer promising solutions to complex software engineering tasks. However, while much research has evaluated code generated by large language models (LLMs), comprehensive studies on agent-generated patches, particularly in real-world settings, are lacking. This study addresses that gap by evaluating 4,892 patches from 10 top-ranked agents on 500 real-world GitHub issues from SWE-Bench Verified, focusing on their impact on code quality. Our analysis shows no single agent dominated, with 170 issues unresolved, indicating room for improvement. Even for patches that passed unit tests and resolved issues, agents made different file and function modifications compared to the gold patches from repository developers, revealing limitations in the benchmark's test case coverage. Most agents maintained code reliability and security, avoiding new bugs or vulnerabilities; while some agents increased code complexity, many reduced code duplication and minimized code smells. Finally, agents performed better on simpler codebases, suggesting that breaking complex tasks into smaller sub-tasks could improve effectiveness. This study provides the first comprehensive evaluation of agent-generated patches on real-world GitHub issues, offering insights to advance AI-driven software development.

TurboRAG: Accelerating Retrieval-Augmented Generation with Precomputed KV Caches for Chunked Text

Current Retrieval-Augmented Generation (RAG) systems concatenate and process numerous retrieved document chunks for prefill which requires a large volume of computation, therefore leading to significant latency in time-to-first-token (TTFT). To reduce the computation overhead as well as TTFT, we introduce TurboRAG, a novel RAG system that redesigns the inference paradigm of the current RAG system by first pre-computing and storing the key-value (KV) caches of documents offline, and then directly retrieving the saved KV cache for prefill. Hence, online computation of KV caches is eliminated during inference. In addition, we provide a number of insights into the mask matrix and positional embedding mechanisms, plus fine-tune a pretrained language model to maintain model accuracy of TurboRAG. Our approach is applicable to most existing large language models and their applications without any requirement in modification of models and inference systems. Experimental results across a suite of RAG benchmarks demonstrate that TurboRAG reduces TTFT by up to 9.4x compared to the conventional RAG systems (on an average of 8.6x), but reserving comparable performance to the standard RAG systems.

TimeCNN: Refining Cross-Variable Interaction on Time Point for Time Series Forecasting

Time series forecasting is extensively applied across diverse domains. Transformer-based models demonstrate significant potential in modeling cross-time and cross-variable interaction. However, we notice that the cross-variable correlation of multivariate time series demonstrates multifaceted (positive and negative correlations) and dynamic progression over time, which is not well captured by existing Transformer-based models. To address this issue, we propose a TimeCNN model to refine cross-variable interactions to enhance time series forecasting. Its key innovation is timepoint-independent, where each time point has an independent convolution kernel, allowing each time point to have its independent model to capture relationships among variables. This approach effectively handles both positive and negative correlations and adapts to the evolving nature of variable relationships over time. Extensive experiments conducted on 12 real-world datasets demonstrate that TimeCNN consistently outperforms state-of-the-art models. Notably, our model achieves significant reductions in computational requirements (approximately 60.46%) and parameter count (about 57.50%), while delivering inference speeds 3 to 4 times faster than the benchmark iTransformer model

Promise and Peril of Collaborative Code Generation Models: Balancing Effectiveness and Memorization

In the rapidly evolving field of machine learning, training models with datasets from various locations and organizations presents significant challenges due to privacy and legal concerns. The exploration of effective collaborative training settings capable of leveraging valuable knowledge from distributed and isolated datasets is increasingly crucial. This study investigates key factors that impact the effectiveness of collaborative training methods in code next-token prediction, as well as the correctness and utility of the generated code, demonstrating the promise of such methods. Additionally, we evaluate the memorization of different participant training data across various collaborative training settings, including centralized, federated, and incremental training, highlighting their potential risks in leaking data. Our findings indicate that the size and diversity of code datasets are pivotal factors influencing the success of collaboratively trained code models. We show that federated learning achieves competitive performance compared to centralized training while offering better data protection, as evidenced by lower memorization ratios in the generated code. However, federated learning can still produce verbatim code snippets from hidden training data, potentially violating privacy or copyright. Our study further explores effectiveness and memorization patterns in incremental learning, emphasizing the sequence in which individual participant datasets are introduced. We also identify cross-organizational clones as a prevalent challenge in both centralized and federated learning scenarios. Our findings highlight the persistent risk of data leakage during inference, even when training data remains unseen. We conclude with recommendations for practitioners and researchers to optimize multisource datasets, propelling cross-organizational collaboration forward.

CF-PRNet: Coarse-to-Fine Prototype Refining Network for Point Cloud Completion and Reconstruction

In modern agriculture, precise monitoring of plants and fruits is crucial for tasks such as high-throughput phenotyping and automated harvesting. This paper addresses the challenge of reconstructing accurate 3D shapes of fruits from partial views, which is common in agricultural settings. We introduce CF-PRNet, a coarse-to-fine prototype refining network, leverages high-resolution 3D data during the training phase but requires only a single RGB-D image for real-time inference. Our approach begins by extracting the incomplete point cloud data that constructed from a partial view of a fruit with a series of convolutional blocks. The extracted features inform the generation of scaling vectors that refine two sequentially constructed 3D mesh prototypes - one coarse and one fine-grained. This progressive refinement facilitates the detailed completion of the final point clouds, achieving detailed and accurate reconstructions. CF-PRNet demonstrates excellent performance metrics with a Chamfer Distance of 3.78, an F1 Score of 66.76%, a Precision of 56.56%, and a Recall of 85.31%, and win the first place in the Shape Completion and Reconstruction of Sweet Peppers Challenge.