Speculative decoding accelerates large language model (LLM) inference by using a lightweight drafter to propose multiple tokens, which are then verified in parallel by the base model. While effective in English, existing methods often struggle in multilingual scenarios due to static vocabularies and the lack of language-specific instruction data. To address these limitations, we present AdaSpec, a multilingual speculative decoding framework that dynamically adapts both the drafter and vocabulary at decoding time. AdaSpec generates language-specific instruction data using the LLM itself, enabling training of drafters for low-resource languages. It also constructs adaptive vocabularies tailored to each language's characteristics. In addition, we introduce Multi-SpecBench, a comprehensive multilingual benchmark covering seven languages and seven generation tasks, to evaluate multilingual speculative decoding performance. Extensive experiments show that AdaSpec achieves up to 2.3× speedup over the state-of-the-art method of EAGLE-2, even in English, demonstrating its effectiveness across diverse languages and tasks.

Speculative decoding accelerates large language model (LLM) inference through a draftand-verify paradigm, yet existing methods face three key limitations: reliance on fixed draft templates that ignore device-specific verification costs, lack of mechanisms to assess draft token quality, and suboptimal tree expansion strategies.We introduce UNISPEC, a trainingfree, lossless speculative decoding framework that enables robust, plug-and-play LLM acceleration across diverse hardware configurations and languages.UNISPEC incorporates three novel components: (1) a device-aware calibration mechanism that determines the optimal draft size by measuring the acceptancetime trade-off on each target device; (2) a confidence score estimation module that assigns quality scores to n-grams based on the verifier's token probabilities, enabling selective retention of high-quality draft candidates; and (3) an improved tree expansion strategy that broadens first-level exploration and applies threshold-based filtering to prune lowconfidence nodes.To comprehensively evaluate multilingual performance, we create a comprehensive benchmark, covering seven languages across seven generation tasks.Experiments with various LLM architectures, hardware environments, and languages demonstrate that UNISPEC consistently outperforms existing training-free methods, achieving speedups of up to 2.6× while maintaining output quality identical to standard autoregressive decoding.Our code and benchmark are publicly available.

Inference with modern Large Language Models (LLMs) is both computationally expensive and time-consuming.Speculative decoding has emerged as a promising solution, but existing approaches face key limitations: training-based methods require a draft model that is challenging to obtain and lacks generalizability, while training-free methods offer limited speedup gains.In this work, we present SPECTRA, a novel framework for accelerating LLM inference without the need for additional training or modification to the original LLM.SPECTRA introduces two new techniques for efficiently utilizing internal and external speculation, each outperforming corresponding state-of-the-art (SOTA) methods independently.When combined, these techniques achieve up to a 4.08x speedup across various benchmarks and LLM architectures, significantly surpassing existing training-free approaches.The implementation of SPECTRA is publicly available.

Large language models (LLMs) and Retrieval-Augmented-Generation (RAG) show remarkable capabilities in Open-domain question-answering (ODQA). Despite the advancements, LLMs tend to generate verbose responses, of which only a small part is the answer phrase. Although the ability to produce the confidence score for the answer is essential when deploying LLMs in high-risk domains, sequence probabilities obtained from LLMs do not correlate well with the probabilities of correctness and thus fail to represent confidence scores. This study introduces Answer-prefix Generation (Anspre) to improve generation quality, allowing the LLMs to output answer phrases and produce highly reliable confidence scores. We guide the model in predicting the answer phrase using an answer prefix and design a ranking score that integrates parametric and non-parametric knowledge. The answer phrases and their corresponding scores enable Anspre to aggregate results from different documents and samplings to boost performance and produce confidence scores highly correlated with correctness. We show that Anspre can be applied to any LLM and present an approach called Self-Anspre to combine Anspre with Self-reflective RAG, a state-of-the-art framework based on reflection tokens. Empirical evaluation on popular ODQA benchmarks shows that Anspre and Self-Anspre significantly improve state-of-the-art LLMs and RAG frameworks. An in-depth analysis shows that confidence scores produced by Anspre are highly correlated to the likelihood of correctness.

Face-swap models have increasingly gained popularity in recent years because of their improvement in generation quality and applications in privacy protection and entertainment. The study primarily examines underlying factors that influence individuals’ intention to purchase Face swap apps. A research framework drawing upon the social comparison theory was proposed to explain the relationship between four categories of social comparison (upward identification, upward contrast, downward identification, downward contrast) on self-esteem, appearance dissatisfaction, curiosity, perceived enjoyment, privacy concern, and purchase intention toward paid apps. The PLS-SEM technique was employed to analyze data from a survey of 315 respondents. The results indicated that upon the aforementioned variables, curiosity and privacy concerns were two key factors driving users a higher intention to purchase Face-swap apps. Meanwhile, perceived enjoyment had no significant impact on purchase intention. These exploratory findings contributed a number of theoretical and practical implications.