Hydragen: High-Throughput LLM Inference with Shared Prefixes

Transformer-based large language models (LLMs) are now deployed to hundreds of millions of users. LLM inference is commonly performed on batches of sequences that share a prefix, such as few-shot examples or a chatbot system prompt. Decoding in this large-batch setting can be bottlenecked by the attention operation, which reads large key-value (KV) caches from memory and computes inefficient matrix-vector products for every sequence in the batch. In this work, we introduce Hydragen, a hardware-aware exact implementation of attention with shared prefixes. Hydragen computes attention over the shared prefix and unique suffixes separately. This decomposition enables efficient prefix attention by batching queries together across sequences, reducing redundant memory reads and enabling the use of hardware-friendly matrix multiplications. Our method can improve end-to-end LLM throughput by up to 32x against competitive baselines, with speedup growing with the batch size and shared prefix length. Hydragen also enables the use of very long shared contexts: with a high batch size, increasing the prefix length from 1K to 16K tokens decreases Hydragen throughput by less than 15%, while the throughput of baselines drops by over 90%. Hydragen generalizes beyond simple prefix-suffix decomposition and can be applied to tree-based prompt sharing patterns, allowing us to further reduce inference time on competitive programming problems by 55%.

Progressive Conservative Adaptation for Evolving Target Domains

Conventional domain adaptation typically transfers knowledge from a source domain to a stationary target domain. However, in many real-world cases, target data usually emerge sequentially and have continuously evolving distributions. Restoring and adapting to such target data results in escalating computational and resource consumption over time. Hence, it is vital to devise algorithms to address the evolving domain adaptation (EDA) problem, \emph{i.e.,} adapting models to evolving target domains without access to historic target domains. To achieve this goal, we propose a simple yet effective approach, termed progressive conservative adaptation (PCAda). To manage new target data that diverges from previous distributions, we fine-tune the classifier head based on the progressively updated class prototypes. Moreover, as adjusting to the most recent target domain can interfere with the features learned from previous target domains, we develop a conservative sparse attention mechanism. This mechanism restricts feature adaptation within essential dimensions, thus easing the inference related to historical knowledge. The proposed PCAda is implemented with a meta-learning framework, which achieves the fast adaptation of the classifier with the help of the progressively updated class prototypes in the inner loop and learns a generalized feature without severely interfering with the historic knowledge via the conservative sparse attention in the outer loop. Experiments on Rotated MNIST, Caltran, and Portraits datasets demonstrate the effectiveness of our method.

No-Regret Reinforcement Learning in Smooth MDPs

Obtaining no-regret guarantees for reinforcement learning (RL) in the case of problems with continuous state and/or action spaces is still one of the major open challenges in the field. Recently, a variety of solutions have been proposed, but besides very specific settings, the general problem remains unsolved. In this paper, we introduce a novel structural assumption on the Markov decision processes (MDPs), namely $\nu-$smoothness, that generalizes most of the settings proposed so far (e.g., linear MDPs and Lipschitz MDPs). To face this challenging scenario, we propose two algorithms for regret minimization in $\nu-$smooth MDPs. Both algorithms build upon the idea of constructing an MDP representation through an orthogonal feature map based on Legendre polynomials. The first algorithm, \textsc{Legendre-Eleanor}, archives the no-regret property under weaker assumptions but is computationally inefficient, whereas the second one, \textsc{Legendre-LSVI}, runs in polynomial time, although for a smaller class of problems. After analyzing their regret properties, we compare our results with state-of-the-art ones from RL theory, showing that our algorithms achieve the best guarantees.

Human Observation-Inspired Trajectory Prediction for Autonomous Driving in Mixed-Autonomy Traffic Environments

In the burgeoning field of autonomous vehicles (AVs), trajectory prediction remains a formidable challenge, especially in mixed autonomy environments. Traditional approaches often rely on computational methods such as time-series analysis. Our research diverges significantly by adopting an interdisciplinary approach that integrates principles of human cognition and observational behavior into trajectory prediction models for AVs. We introduce a novel "adaptive visual sector" mechanism that mimics the dynamic allocation of attention human drivers exhibit based on factors like spatial orientation, proximity, and driving speed. Additionally, we develop a "dynamic traffic graph" using Convolutional Neural Networks (CNN) and Graph Attention Networks (GAT) to capture spatio-temporal dependencies among agents. Benchmark tests on the NGSIM, HighD, and MoCAD datasets reveal that our model (GAVA) outperforms state-of-the-art baselines by at least 15.2%, 19.4%, and 12.0%, respectively. Our findings underscore the potential of leveraging human cognition principles to enhance the proficiency and adaptability of trajectory prediction algorithms in AVs. The code for the proposed model is available at our Github.

INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection

Knowledge hallucination have raised widespread concerns for the security and reliability of deployed LLMs. Previous efforts in detecting hallucinations have been employed at logit-level uncertainty estimation or language-level self-consistency evaluation, where the semantic information is inevitably lost during the token-decoding procedure. Thus, we propose to explore the dense semantic information retained within LLMs' \textbf{IN}ternal \textbf{S}tates for halluc\textbf{I}nation \textbf{DE}tection (\textbf{INSIDE}). In particular, a simple yet effective \textbf{EigenScore} metric is proposed to better evaluate responses' self-consistency, which exploits the eigenvalues of responses' covariance matrix to measure the semantic consistency/diversity in the dense embedding space. Furthermore, from the perspective of self-consistent hallucination detection, a test time feature clipping approach is explored to truncate extreme activations in the internal states, which reduces overconfident generations and potentially benefits the detection of overconfident hallucinations. Extensive experiments and ablation studies are performed on several popular LLMs and question-answering (QA) benchmarks, showing the effectiveness of our proposal.

Enhancing LLM-Based Coding Tools through Native Integration of IDE-Derived Static Context

Large Language Models (LLMs) have achieved remarkable success in code completion, as evidenced by their essential roles in developing code assistant services such as Copilot. Being trained on in-file contexts, current LLMs are quite effective in completing code for single source files. However, it is challenging for them to conduct repository-level code completion for large software projects that require cross-file information. Existing research on LLM-based repository-level code completion identifies and integrates cross-file contexts, but it suffers from low accuracy and limited context length of LLMs. In this paper, we argue that Integrated Development Environments (IDEs) can provide direct, accurate and real-time cross-file information for repository-level code completion. We propose IDECoder, a practical framework that leverages IDE native static contexts for cross-context construction and diagnosis results for self-refinement. IDECoder utilizes the rich cross-context information available in IDEs to enhance the capabilities of LLMs of repository-level code completion. We conducted preliminary experiments to validate the performance of IDECoder and observed that this synergy represents a promising trend for future exploration.

Entropy-regularized Diffusion Policy with Q-Ensembles for Offline Reinforcement Learning

This paper presents advanced techniques of training diffusion policies for offline reinforcement learning (RL). At the core is a mean-reverting stochastic differential equation (SDE) that transfers a complex action distribution into a standard Gaussian and then samples actions conditioned on the environment state with a corresponding reverse-time SDE, like a typical diffusion policy. We show that such an SDE has a solution that we can use to calculate the log probability of the policy, yielding an entropy regularizer that improves the exploration of offline datasets. To mitigate the impact of inaccurate value functions from out-of-distribution data points, we further propose to learn the lower confidence bound of Q-ensembles for more robust policy improvement. By combining the entropy-regularized diffusion policy with Q-ensembles in offline RL, our method achieves state-of-the-art performance on most tasks in D4RL benchmarks. Code is available at \href{https://github.com/ruoqizzz/Entropy-Regularized-Diffusion-Policy-with-QEnsemble}{https://github.com/ruoqizzz/Entropy-Regularized-Diffusion-Policy-with-QEnsemble}.

SGS-SLAM: Semantic Gaussian Splatting For Neural Dense SLAM

Semantic understanding plays a crucial role in Dense Simultaneous Localization and Mapping (SLAM), facilitating comprehensive scene interpretation. Recent advancements that integrate Gaussian Splatting into SLAM systems have demonstrated its effectiveness in generating high-quality renderings through the use of explicit 3D Gaussian representations. Building on this progress, we propose SGS-SLAM, the first semantic dense visual SLAM system grounded in 3D Gaussians, which provides precise 3D semantic segmentation alongside high-fidelity reconstructions. Specifically, we propose to employ multi-channel optimization during the mapping process, integrating appearance, geometric, and semantic constraints with key-frame optimization to enhance reconstruction quality. Extensive experiments demonstrate that SGS-SLAM delivers state-of-the-art performance in camera pose estimation, map reconstruction, and semantic segmentation, outperforming existing methods meanwhile preserving real-time rendering ability.

Non-Stationary Latent Auto-Regressive Bandits

We consider the stochastic multi-armed bandit problem with non-stationary rewards. We present a novel formulation of non-stationarity in the environment where changes in the mean reward of the arms over time are due to some unknown, latent, auto-regressive (AR) state of order $k$. We call this new environment the latent AR bandit. Different forms of the latent AR bandit appear in many real-world settings, especially in emerging scientific fields such as behavioral health or education where there are few mechanistic models of the environment. If the AR order $k$ is known, we propose an algorithm that achieves $\tilde{O}(k\sqrt{T})$ regret in this setting. Empirically, our algorithm outperforms standard UCB across multiple non-stationary environments, even if $k$ is mis-specified.

Focal Modulation Networks for Interpretable Sound Classification

The increasing success of deep neural networks has raised concerns about their inherent black-box nature, posing challenges related to interpretability and trust. While there has been extensive exploration of interpretation techniques in vision and language, interpretability in the audio domain has received limited attention, primarily focusing on post-hoc explanations. This paper addresses the problem of interpretability by-design in the audio domain by utilizing the recently proposed attention-free focal modulation networks (FocalNets). We apply FocalNets to the task of environmental sound classification for the first time and evaluate their interpretability properties on the popular ESC-50 dataset. Our method outperforms a similarly sized vision transformer both in terms of accuracy and interpretability. Furthermore, it is competitive against PIQ, a method specifically designed for post-hoc interpretation in the audio domain.