Accelerating the inference of large language models (LLMs) is an important challenge in artificial intelligence. This paper introduces distributed speculative inference (DSI), a novel distributed inference algorithm that is provably faster than speculative inference (SI) [leviathan2023fast, chen2023accelerating, miao2023specinfer] and traditional autoregressive inference (non-SI). Like other SI algorithms, DSI works on frozen LLMs, requiring no training or architectural modifications, and it preserves the target distribution. Prior studies on SI have demonstrated empirical speedups (compared to non-SI) but require a fast and accurate drafter LLM. In practice, off-the-shelf LLMs often do not have matching drafters that are sufficiently fast and accurate. We show a gap: SI gets slower than non-SI when using slower or less accurate drafters. We close this gap by proving that DSI is faster than both SI and non-SI given any drafters. By orchestrating multiple instances of the target and drafters, DSI is not only faster than SI but also supports LLMs that cannot be accelerated with SI. Our simulations show speedups of off-the-shelf LLMs in realistic settings: DSI is 1.29-1.92x faster than SI.
相關內容
This paper investigates the relationships between hyperparameters of machine learning and fairness. Data-driven solutions are increasingly used in critical socio-technical applications where ensuring fairness is important. Rather than explicitly encoding decision logic via control and data structures, the ML developers provide input data, perform some pre-processing, choose ML algorithms, and tune hyperparameters (HPs) to infer a program that encodes the decision logic. Prior works report that the selection of HPs can significantly influence fairness. However, tuning HPs to find an ideal trade-off between accuracy, precision, and fairness has remained an expensive and tedious task. Can we predict fairness of HP configuration for a given dataset? Are the predictions robust to distribution shifts? We focus on group fairness notions and investigate the HP space of 5 training algorithms. We first find that tree regressors and XGBoots significantly outperformed deep neural networks and support vector machines in accurately predicting the fairness of HPs. When predicting the fairness of ML hyperparameters under temporal distribution shift, the tree regressors outperforms the other algorithms with reasonable accuracy. However, the precision depends on the ML training algorithm, dataset, and protected attributes. For example, the tree regressor model was robust for training data shift from 2014 to 2018 on logistic regression and discriminant analysis HPs with sex as the protected attribute; but not for race and other training algorithms. Our method provides a sound framework to efficiently perform fine-tuning of ML training algorithms and understand the relationships between HPs and fairness.
The significant advancements in large language models (LLMs) give rise to a promising research direction, i.e., leveraging LLMs as recommenders (LLMRec). The efficacy of LLMRec arises from the open-world knowledge and reasoning capabilities inherent in LLMs. LLMRec acquires the recommendation capabilities through instruction tuning based on user interaction data. However, in order to protect user privacy and optimize utility, it is also crucial for LLMRec to intentionally forget specific user data, which is generally referred to as recommendation unlearning. In the era of LLMs, recommendation unlearning poses new challenges for LLMRec in terms of \textit{inefficiency} and \textit{ineffectiveness}. Existing unlearning methods require updating billions of parameters in LLMRec, which is costly and time-consuming. Besides, they always impact the model utility during the unlearning process. To this end, we propose \textbf{E2URec}, the first \underline{E}fficient and \underline{E}ffective \underline{U}nlearning method for LLM\underline{Rec}. Our proposed E2URec enhances the unlearning efficiency by updating only a few additional LoRA parameters, and improves the unlearning effectiveness by employing a teacher-student framework, where we maintain multiple teacher networks to guide the unlearning process. Extensive experiments show that E2URec outperforms state-of-the-art baselines on two real-world datasets. Specifically, E2URec can efficiently forget specific data without affecting recommendation performance. The source code is at \url{//github.com/justarter/E2URec}.
The interactive use of large language models (LLMs) in AI assistants (at work, home, etc.) introduces a new set of inference-time privacy risks: LLMs are fed different types of information from multiple sources in their inputs and are expected to reason about what to share in their outputs, for what purpose and with whom, within a given context. In this work, we draw attention to the highly critical yet overlooked notion of contextual privacy by proposing ConfAIde, a benchmark designed to identify critical weaknesses in the privacy reasoning capabilities of instruction-tuned LLMs. Our experiments show that even the most capable models such as GPT-4 and ChatGPT reveal private information in contexts that humans would not, 39% and 57% of the time, respectively. This leakage persists even when we employ privacy-inducing prompts or chain-of-thought reasoning. Our work underscores the immediate need to explore novel inference-time privacy-preserving approaches, based on reasoning and theory of mind.
Statistical modelling in the presence of data organized in groups is a crucial task in Bayesian statistics. The present paper conceives a mixture model based on a novel family of Bayesian priors designed for multilevel data and obtained by normalizing a finite point process. In particular, the work extends the popular Mixture of Finite Mixture model to the hierarchical framework to capture heterogeneity within and between groups. A full distribution theory for this new family and the induced clustering is developed, including the marginal, posterior, and predictive distributions. Efficient marginal and conditional Gibbs samplers are designed to provide posterior inference. The proposed mixture model overcomes the Hierarchical Dirichlet Process, the utmost tool for handling multilevel data, in terms of analytical feasibility, clustering discovery, and computational time. The motivating application comes from the analysis of shot put data, which contains performance measurements of athletes across different seasons. In this setting, the proposed model is exploited to induce clustering of the observations across seasons and athletes. By linking clusters across seasons, similarities and differences in athletes' performances are identified.
Minimal pairs are a well-established approach to evaluating the grammatical knowledge of language models. However, existing resources for minimal pairs address a limited number of languages and lack diversity of language-specific grammatical phenomena. This paper introduces the Russian Benchmark of Linguistic Minimal Pairs (RuBLiMP), which includes 45k pairs of sentences that differ in grammaticality and isolate a morphological, syntactic, or semantic phenomenon. In contrast to existing benchmarks of linguistic minimal pairs, RuBLiMP is created by applying linguistic perturbations to automatically annotated sentences from open text corpora and carefully curating test data. We describe the data collection protocol and present the results of evaluating 25 language models in various scenarios. We find that the widely used language models for Russian are sensitive to morphological and agreement-oriented contrasts but fall behind humans on phenomena requiring understanding of structural relations, negation, transitivity, and tense. RuBLiMP, the codebase, and other materials are publicly available.
In the field of large language models (LLMs), Knowledge Distillation (KD) is a critical technique for transferring capabilities from teacher models to student models. However, existing KD methods face limitations and challenges in distillation of LLMs, including efficiency and insufficient measurement capabilities of traditional KL divergence. It is shown that LLMs can serve as an implicit reward function, which we define as a supplement to KL divergence. In this work, we propose Direct Preference Knowledge Distillation (DPKD) for LLMs. DPKD utilizes distribution divergence to represent the preference loss and implicit reward function. We re-formulate KD of LLMs into two stages: first optimizing and objective consisting of implicit reward and reverse KL divergence and then improving the preference probability of teacher outputs over student outputs. We conducted experiments and analysis on various datasets with LLM parameters ranging from 120M to 13B and demonstrate the broad applicability and effectiveness of our DPKD approach. Meanwhile, we prove the value and effectiveness of the introduced implicit reward and output preference in KD through experiments and theoretical analysis. The DPKD method outperforms the baseline method in both output response precision and exact match percentage. Code and data are available at //aka.ms/dpkd.
As large language models (LLMs) become more capable, fine-tuning techniques for aligning with human intent are increasingly important. A key consideration for aligning these models is how to most effectively use human resources, or model resources in the case where LLMs themselves are used as oracles. Reinforcement learning from Human or AI preferences (RLHF/RLAIF) is the most prominent example of such a technique, but is complex and often unstable. Direct Preference Optimization (DPO) has recently been proposed as a simpler and more stable alternative. In this work, we develop an active learning strategy for DPO to make better use of preference labels. We propose a practical acquisition function for prompt/completion pairs based on the predictive entropy of the language model and a measure of certainty of the implicit preference model optimized by DPO. We demonstrate how our approach improves both the rate of learning and final performance of fine-tuning on pairwise preference data.
In this paper, the interference cancellation information geometry approaches (IC-IGAs) for massive MIMO channel estimation are proposed. The proposed algorithms are low-complexity approximations of the minimum mean square error (MMSE) estimation. To illustrate the proposed algorithms, a unified framework of the information geometry approach for channel estimation and its geometric explanation are described first. Then, a modified form that has the same mean as the MMSE estimation is constructed. Based on this, the IC-IGA algorithm and the interference cancellation simplified information geometry approach (IC-SIGA) are derived by applying the information geometry framework. The a posteriori means on the equilibrium of the proposed algorithms are proved to be equal to the mean of MMSE estimation, and the complexity of the IC-SIGA algorithm in practical massive MIMO systems is further reduced by considering the beam-based statistical channel model (BSCM) and fast Fourier transform (FFT). Simulation results show that the proposed methods achieve similar performance as the existing information geometry approach (IGA) with lower complexity.
The advent of large language models marks a revolutionary breakthrough in artificial intelligence. With the unprecedented scale of training and model parameters, the capability of large language models has been dramatically improved, leading to human-like performances in understanding, language synthesizing, and common-sense reasoning, etc. Such a major leap-forward in general AI capacity will change the pattern of how personalization is conducted. For one thing, it will reform the way of interaction between humans and personalization systems. Instead of being a passive medium of information filtering, large language models present the foundation for active user engagement. On top of such a new foundation, user requests can be proactively explored, and user's required information can be delivered in a natural and explainable way. For another thing, it will also considerably expand the scope of personalization, making it grow from the sole function of collecting personalized information to the compound function of providing personalized services. By leveraging large language models as general-purpose interface, the personalization systems may compile user requests into plans, calls the functions of external tools to execute the plans, and integrate the tools' outputs to complete the end-to-end personalization tasks. Today, large language models are still being developed, whereas the application in personalization is largely unexplored. Therefore, we consider it to be the right time to review the challenges in personalization and the opportunities to address them with LLMs. In particular, we dedicate this perspective paper to the discussion of the following aspects: the development and challenges for the existing personalization system, the newly emerged capabilities of large language models, and the potential ways of making use of large language models for personalization.
As artificial intelligence (AI) models continue to scale up, they are becoming more capable and integrated into various forms of decision-making systems. For models involved in moral decision-making, also known as artificial moral agents (AMA), interpretability provides a way to trust and understand the agent's internal reasoning mechanisms for effective use and error correction. In this paper, we provide an overview of this rapidly-evolving sub-field of AI interpretability, introduce the concept of the Minimum Level of Interpretability (MLI) and recommend an MLI for various types of agents, to aid their safe deployment in real-world settings.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191