Large Language models (LLMs) have become a research hotspot. To accelerate the inference of LLMs, storing computed caches in memory has become the standard technique. However, as the inference length increases, growing KV caches might lead to out-of-memory issues. Many existing methods address this issue through KV cache compression, primarily by preserving key tokens throughout all layers to reduce information loss. Most of them allocate a uniform budget size for each layer to retain. However, we observe that the minimum budget sizes needed to retain essential information vary across layers and models based on the perspectives of attention and hidden state output. Building on this observation, this paper proposes a simple yet effective KV cache compression method that leverages layer uncertainty to allocate budget size for each layer. Experimental results show that the proposed method can reduce memory usage of the KV caches to only ~20% when compared to full KV inference while achieving nearly lossless performance.

Enabling LLMs to handle lengthy context is currently a research hotspot. Most LLMs are built upon rotary position embedding (RoPE), a popular position encoding method. Therefore, a prominent path is to extrapolate the RoPE trained on comparably short texts to far longer texts. A heavy bunch of efforts have been dedicated to boosting the extrapolation via extending the formulations of the RoPE, however, few of them have attempted to showcase their inner workings comprehensively. In this paper, we are driven to offer a straightforward yet in-depth understanding of RoPE extensions from an attention perspective and on two benchmarking tasks. A broad array of experiments reveals several valuable findings: 1) Maintaining attention patterns to those at the pretrained length improves extrapolation; 2) Large attention uncertainty leads to retrieval errors; 3) Using longer continual pretraining lengths for RoPE extensions could reduce attention uncertainty and significantly enhance extrapolation.

Agentic Retrieval-Augmented Generation (Agentic RAG) enhances the processing capability for complex tasks through dynamic retrieval and adaptive workflows. Recent advances (e.g., Search-R1) have shown that outcome-supervised reinforcement learning demonstrate strong performance. However, this approach still suffers from inefficient exploration, sparse reward signals, and ambiguous global reward feedback.To address these challenges, we propose DecEx-RAG, which models RAG as a Markov Decision Process (MDP) incorporating decision-making and execution, while introducing an efficient pruning strategy to optimize data expansion. Through comprehensive process-level policy optimization, DecEx-RAG significantly enhances the autonomous task decomposition, dynamic retrieval, and high-quality answer generation capabilities of large language models (LLMs). Experiments show that DecEx-RAG achieves an average absolute performance improvement of 6.2% across six datasets, significantly outperforming existing baselines. Moreover, the pruning strategy improves data construction efficiency by nearly 6 ×, providing an efficient solution for process-supervised RAG training. The code is available at https://github.com/sdsxdxl/DecEx-RAG.

Large language models usually suffer from multiple-file coding scenarios where strong inter-file dependencies manifest, typically demonstrated in SWE-bench. To mitigate this issue, we propose Think-Search-Patch (TSP), a retrieval-augmented reasoning framework for repository-level code repair. At the Think stage, our system breaks down a coding task and creates clear search query. Next, at the Search stage, it retrieves relevant code snippets using models like E5. At the final Patch stage, it generates standardized patches based on the key snippets. In addition the proposed framework, we enhance system reliability through a two-stage training process. At the first stage, the system undergoes supervised fine-tuning (SFT) on our TSP dataset. At the subsequent stage, we employ rejection sampling with correction to generate preference pairs for Direct Preference Optimization (DPO) training, thereby reducing errors in the intermediate phases. Experimental results demonstrate that TSP framework enhances retrieval accuracy and repair success on SWE-bench Lite, even surpassing models with a larger size in managing extensive code contexts and successfully addressing bugs spanning across multiple files. All data and code available at https://github.com/Gengar0215/TSP-framework.

Long-context efficiency has recently become a trending topic in serving large language models (LLMs). And mixture of depths (MoD) is proposed as a perfect fit to bring down both latency and memory. In this paper, however, we discover that MoD can barely transform existing LLMs without costly training over an extensive number of tokens. To enable the transformations from any LLMs to MoD ones, we showcase top-k operator in MoD should be promoted to threshold-p operator, and refinement to architecture and data should also be crafted along. All these designs form our method termed MoDification. Through a comprehensive set of experiments covering model scales from 3B to 70B, we exhibit MoDification strikes an excellent balance between efficiency and effectiveness. MoDification can achieve up to ~1.2× speedup in latency and ~1.8× reduction in memory compared to original LLMs especially in long-context applications.

Simultaneous Machine Translation (SiMT) aims to yield a real-time partial translation with a monotonically growing source-side context.However, there is a counterintuitive phenomenon about the context usage between training and inference: *e.g.*, in wait-k inference, model consistently trained with wait-k is much worse than that model inconsistently trained with wait-k' (k'≠ k) in terms of translation quality. To this end, we first investigate the underlying reasons behind this phenomenon and uncover the following two factors: 1) the limited correlation between translation quality and training loss; 2) exposure bias between training and inference. Based on both reasons, we then propose an effective training approach called context consistency training accordingly, which encourages consistent context usage between training and inference by optimizing translation quality and latency as bi-objectives and exposing the predictions to the model during the training. The experiments on three language pairs demonstrate that our SiMT system encouraging context consistency outperforms existing SiMT systems with context inconsistency for the first time.

It is widely known that hallucination is a critical issue in Simultaneous Machine Translation (SiMT) due to the absence of source-side information. While many efforts have been made to enhance performance for SiMT, few of them attempt to understand and analyze hallucination in SiMT.Therefore, we conduct a comprehensive analysis of hallucination in SiMT from two perspectives: understanding the distribution of hallucination words and the target-side context usage of them.Intensive experiments demonstrate some valuable findings and particularly show that it is possible to alleviate hallucination by decreasing the over usage of target-side information for SiMT.