GeoERM: Geometry-Aware Multi-Task Representation Learning on Riemannian Manifolds

Multi-Task Learning (MTL) seeks to boost statistical power and learning efficiency by discovering structure shared across related tasks. State-of-the-art MTL representation methods, however, usually treat the latent representation matrix as a point in ordinary Euclidean space, ignoring its often non-Euclidean geometry, thus sacrificing robustness when tasks are heterogeneous or even adversarial. We propose GeoERM, a geometry-aware MTL framework that embeds the shared representation on its natural Riemannian manifold and optimizes it via explicit manifold operations. Each training cycle performs (i) a Riemannian gradient step that respects the intrinsic curvature of the search space, followed by (ii) an efficient polar retraction to remain on the manifold, guaranteeing geometric fidelity at every iteration. The procedure applies to a broad class of matrix-factorized MTL models and retains the same per-iteration cost as Euclidean baselines. Across a set of synthetic experiments with task heterogeneity and on a wearable-sensor activity-recognition benchmark, GeoERM consistently improves estimation accuracy, reduces negative transfer, and remains stable under adversarial label noise, outperforming leading MTL and single-task alternatives.

LLaMA-Omni2: LLM-based Real-time Spoken Chatbot with Autoregressive Streaming Speech Synthesis

Real-time, intelligent, and natural speech interaction is an essential part of the next-generation human-computer interaction. Recent advancements have showcased the potential of building intelligent spoken chatbots based on large language models (LLMs). In this paper, we introduce LLaMA-Omni 2, a series of speech language models (SpeechLMs) ranging from 0.5B to 14B parameters, capable of achieving high-quality real-time speech interaction. LLaMA-Omni 2 is built upon the Qwen2.5 series models, integrating a speech encoder and an autoregressive streaming speech decoder. Despite being trained on only 200K multi-turn speech dialogue samples, LLaMA-Omni 2 demonstrates strong performance on several spoken question answering and speech instruction following benchmarks, surpassing previous state-of-the-art SpeechLMs like GLM-4-Voice, which was trained on millions of hours of speech data.

Persona-judge: Personalized Alignment of Large Language Models via Token-level Self-judgment

Aligning language models with human preferences presents significant challenges, particularly in achieving personalization without incurring excessive computational costs. Existing methods rely on reward signals and additional annotated data, limiting their scalability and adaptability to diverse human values. To address these challenges, we introduce Persona-judge, a novel discriminative paradigm that enables training-free personalized alignment with unseen preferences. Instead of optimizing policy parameters through external reward feedback, Persona-judge leverages the intrinsic preference judgment capabilities of the model. Specifically, a draft model generates candidate tokens conditioned on a given preference, while a judge model, embodying another preference, cross-validates the predicted tokens whether to be accepted. Experimental results demonstrate that Persona-judge, using the inherent preference evaluation mechanisms of the model, offers a scalable and computationally efficient solution to personalized alignment, paving the way for more adaptive customized alignment.

StruPhantom: Evolutionary Injection Attacks on Black-Box Tabular Agents Powered by Large Language Models

The proliferation of autonomous agents powered by large language models (LLMs) has revolutionized popular business applications dealing with tabular data, i.e., tabular agents. Although LLMs are observed to be vulnerable against prompt injection attacks from external data sources, tabular agents impose strict data formats and predefined rules on the attacker's payload, which are ineffective unless the agent navigates multiple layers of structural data to incorporate the payload. To address the challenge, we present a novel attack termed StruPhantom which specifically targets black-box LLM-powered tabular agents. Our attack designs an evolutionary optimization procedure which continually refines attack payloads via the proposed constrained Monte Carlo Tree Search augmented by an off-topic evaluator. StruPhantom helps systematically explore and exploit the weaknesses of target applications to achieve goal hijacking. Our evaluation validates the effectiveness of StruPhantom across various LLM-based agents, including those on real-world platforms, and attack scenarios. Our attack achieves over 50% higher success rates than baselines in enforcing the application's response to contain phishing links or malicious codes.

LevelRAG: Enhancing Retrieval-Augmented Generation with Multi-hop Logic Planning over Rewriting Augmented Searchers

Retrieval-Augmented Generation (RAG) is a crucial method for mitigating hallucinations in Large Language Models (LLMs) and integrating external knowledge into their responses. Existing RAG methods typically employ query rewriting to clarify the user intent and manage multi-hop logic, while using hybrid retrieval to expand search scope. However, the tight coupling of query rewriting to the dense retriever limits its compatibility with hybrid retrieval, impeding further RAG performance improvements. To address this challenge, we introduce a high-level searcher that decomposes complex queries into atomic queries, independent of any retriever-specific optimizations. Additionally, to harness the strengths of sparse retrievers for precise keyword retrieval, we have developed a new sparse searcher that employs Lucene syntax to enhance retrieval accuracy.Alongside web and dense searchers, these components seamlessly collaborate within our proposed method, \textbf{LevelRAG}. In LevelRAG, the high-level searcher orchestrates the retrieval logic, while the low-level searchers (sparse, web, and dense) refine the queries for optimal retrieval. This approach enhances both the completeness and accuracy of the retrieval process, overcoming challenges associated with current query rewriting techniques in hybrid retrieval scenarios. Empirical experiments conducted on five datasets, encompassing both single-hop and multi-hop question answering tasks, demonstrate the superior performance of LevelRAG compared to existing RAG methods. Notably, LevelRAG outperforms the state-of-the-art proprietary model, GPT4o, underscoring its effectiveness and potential impact on the RAG field.

LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Token

The advent of real-time large multimodal models (LMMs) like GPT-4o has sparked considerable interest in efficient LMMs. LMM frameworks typically encode visual inputs into vision tokens (continuous representations) and integrate them and textual instructions into the context of large language models (LLMs), where large-scale parameters and numerous context tokens (predominantly vision tokens) result in substantial computational overhead. Previous efforts towards efficient LMMs always focus on replacing the LLM backbone with smaller models, while neglecting the crucial issue of token quantity. In this paper, we introduce LLaVA-Mini, an efficient LMM with minimal vision tokens. To achieve a high compression ratio of vision tokens while preserving visual information, we first analyze how LMMs understand vision tokens and find that most vision tokens only play a crucial role in the early layers of LLM backbone, where they mainly fuse visual information into text tokens. Building on this finding, LLaVA-Mini introduces modality pre-fusion to fuse visual information into text tokens in advance, thereby facilitating the extreme compression of vision tokens fed to LLM backbone into one token. LLaVA-Mini is a unified large multimodal model that can support the understanding of images, high-resolution images, and videos in an efficient manner. Experiments across 11 image-based and 7 video-based benchmarks demonstrate that LLaVA-Mini outperforms LLaVA-v1.5 with just 1 vision token instead of 576. Efficiency analyses reveal that LLaVA-Mini can reduce FLOPs by 77%, deliver low-latency responses within 40 milliseconds, and process over 10,000 frames of video on the GPU hardware with 24GB of memory.

Large Language Models Are Read/Write Policy-Makers for Simultaneous Generation

Simultaneous generation models write generation results while reading streaming inputs, necessitating a policy-maker to determine the appropriate output timing. Existing simultaneous generation methods generally adopt the traditional encoder-decoder architecture and learn the generation and policy-making capabilities through complex dynamic programming techniques. Although LLMs excel at text generation, they face challenges in taking on the role of policy-makers through traditional training methods, limiting their exploration in simultaneous generation. To overcome these limitations, we propose a novel LLM-driven Simultaneous Generation (LSG) framework, which allows the off-the-shelf LLM to decide the generation timing and produce output concurrently. Specifically, LSG selects the generation policy that minimizes latency as the baseline policy. Referring to the baseline policy, LSG enables the LLM to devise an improved generation policy that better balances latency and generation quality, and writes generation results accordingly. Experiments on simultaneous translation and streaming automatic speech recognition tasks show that our method can achieve state-of-the-art performance utilizing the open-source LLMs and demonstrate practicality in real-world scenarios.

Auto-RAG: Autonomous Retrieval-Augmented Generation for Large Language Models

Iterative retrieval refers to the process in which the model continuously queries the retriever during generation to enhance the relevance of the retrieved knowledge, thereby improving the performance of Retrieval-Augmented Generation (RAG). Existing work typically employs few-shot prompting or manually constructed rules to implement iterative retrieval. This introduces additional inference overhead and overlooks the remarkable reasoning capabilities of Large Language Models (LLMs). In this paper, we introduce Auto-RAG, an autonomous iterative retrieval model centered on the LLM's powerful decision-making capabilities. Auto-RAG engages in multi-turn dialogues with the retriever, systematically planning retrievals and refining queries to acquire valuable knowledge. This process continues until sufficient external information is gathered, at which point the results are presented to the user. To this end, we develop a method for autonomously synthesizing reasoning-based decision-making instructions in iterative retrieval and fine-tuned the latest open-source LLMs. The experimental results indicate that Auto-RAG is capable of autonomous iterative interaction with the retriever, effectively leveraging the remarkable reasoning and decision-making abilities of LLMs, which lead to outstanding performance across six benchmarks. Further analysis reveals that Auto-RAG can autonomously adjust the number of iterations based on the difficulty of the questions and the utility of the retrieved knowledge, without requiring any human intervention. Moreover, Auto-RAG expresses the iterative retrieval process in natural language, enhancing interpretability while providing users with a more intuitive experience\footnote{Code is available at \url{https://github.com/ictnlp/Auto-RAG}.

Learning Monotonic Attention in Transducer for Streaming Generation

Streaming generation models are increasingly utilized across various fields, with the Transducer architecture being particularly popular in industrial applications. However, its input-synchronous decoding mechanism presents challenges in tasks requiring non-monotonic alignments, such as simultaneous translation, leading to suboptimal performance in these contexts. In this research, we address this issue by tightly integrating Transducer's decoding with the history of input stream via a learnable monotonic attention mechanism. Our approach leverages the forward-backward algorithm to infer the posterior probability of alignments between the predictor states and input timestamps, which is then used to estimate the context representations of monotonic attention in training. This allows Transducer models to adaptively adjust the scope of attention based on their predictions, avoiding the need to enumerate the exponentially large alignment space. Extensive experiments demonstrate that our MonoAttn-Transducer significantly enhances the handling of non-monotonic alignments in streaming generation, offering a robust solution for Transducer-based frameworks to tackle more complex streaming generation tasks.

Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration

Inference acceleration of large language models (LLMs) has been put forward in many application scenarios and speculative decoding has shown its advantage in addressing inference acceleration. Speculative decoding usually introduces a draft model to assist the base LLM where the draft model produces drafts and the base LLM verifies the draft for acceptance or rejection. In this framework, the final inference speed is decided by the decoding speed of the draft model and the acceptance rate of the draft provided by the draft model. Currently the widely used draft models usually generate draft tokens for the next several positions in a non-autoregressive way without considering the correlations between draft tokens. Therefore, it has a high decoding speed but an unsatisfactory acceptance rate. In this paper, we focus on how to improve the performance of the draft model and aim to accelerate inference via a high acceptance rate. To this end, we propose a CTC-based draft model which strengthens the correlations between draft tokens during the draft phase, thereby generating higher-quality draft candidate sequences. Experiment results show that compared to strong baselines, the proposed method can achieve a higher acceptance rate and hence a faster inference speed.