MineDraft: A Framework for Batch Parallel Speculative Decoding
Zhenwei Tang, Arun Verma, Zijian Zhou
et al.
Speculative decoding (SD) accelerates large language model inference by using a smaller draft model to propose draft tokens that are subsequently verified by a larger target model. However, the performance of standard SD is often limited by the strictly sequential execution of these drafting and verification stages. To address this, this paper proposes MineDraft, a batch parallel speculative decoding (PSD) framework designed to effectively hide drafting latency by overlapping it with verification. Our theoretical analysis shows that PSD is substantially more efficient than standard SD. MineDraft realizes the PSD through a novel batch-parallel design that maintains two batches of requests, overlapping drafting for one batch with verification for the other. Our experimental results show significant improvements of MineDraft in both throughput (up to 75%) and end-to-end latency (up to 39%) over standard SD. Furthermore, we have implemented MineDraft as a plugin for vLLM, demonstrating its practicality for production-ready inference systems.
The Philosophy and Physics of Duality
Sebastian De Haro, Jeremy Butterfield
This monograph discusses dualities in physics: what dualities are, their main examples--from quantum mechanics and electrodynamics to statistical mechanics, quantum field theory and string theory--and the philosophical questions they raise. Part I first conceptualises dualities and discusses their main roles and themes, including how they are related to familiar notions like symmetry and interpretation. It also discusses the main simple examples of dualities: position-momentum, wave-particle, electric-magnetic, and Kramers-Wannier dualities. Part II discusses advanced examples and their inter-relations: particle-soliton dualities, electric-magnetic dualities in quantum field theories, dualities in string theory, and gauge-gravity duality. This Part ends with discussions of the hole argument, and how string theory counts the microstates of a black hole. Part III is an in-depth discussion of general philosophical issues on which dualities bear: theoretical equivalence (two theories 'saying the same thing, in different words'), scientific realism and the under-determination of theories by data, theory succession and the M-theory programme, explanation, and scientific understanding. It proposes a view of scientific theories that it dubs 'the geometric view of theories'. The book's treatment of the examples is at the advanced undergraduate and graduate level, starting from elementary and progressing to more advanced examples. The discussions of philosophical topics, such as referential semantics, theoretical equivalence, scientific realism and scientific understanding, are both self-contained and in-depth. Thus the book is aimed at students and researchers with an interest in the physical examples and philosophical questions about dualities, and also in how physics and philosophy can fruitfully interact with each other.
en
physics.hist-ph, cond-mat.stat-mech
Tutorial Proposal: Speculative Decoding for Efficient LLM Inference
Heming Xia, Cunxiao Du, Yongqi Li
et al.
This tutorial presents a comprehensive introduction to Speculative Decoding (SD), an advanced technique for LLM inference acceleration that has garnered significant research interest in recent years. SD is introduced as an innovative decoding paradigm to mitigate the high inference latency stemming from autoregressive decoding in LLMs. At each decoding step, SD efficiently drafts several future tokens and then verifies them in parallel. This approach, unlike traditional autoregressive decoding, facilitates the simultaneous decoding of multiple tokens per step, thereby achieving promising 2x-4x speedups in LLM inference while maintaining original distributions. This tutorial delves into the latest techniques in SD, including draft model architectures and verification strategies. Additionally, it explores the acceleration potential and future research directions in this promising field. We aim for this tutorial to elucidate the current research landscape and offer insights for researchers interested in Speculative Decoding, ultimately contributing to more efficient LLM inference.
Speculative Safety-Aware Decoding
Xuekang Wang, Shengyu Zhu, Xueqi Cheng
Despite extensive efforts to align Large Language Models (LLMs) with human values and safety rules, jailbreak attacks that exploit certain vulnerabilities continuously emerge, highlighting the need to strengthen existing LLMs with additional safety properties to defend against these attacks. However, tuning large models has become increasingly resource intensive and may have difficulty ensuring consistent performance. We introduce Speculative Safety-Aware Decoding (SSD), a lightweight decoding-time approach that equips LLMs with the desired safety property while accelerating inference. We assume that there exists a small language model that possesses this desired property. SSD integrates speculative sampling during decoding and leverages the match ratio between the small and composite models to quantify jailbreak risks. This enables SSD to dynamically switch between decoding schemes to prioritize utility or safety, to handle the challenge of different model capacities. The output token is then sampled from a new distribution that combines the distributions of the original and the small models. Experimental results show that SSD successfully equips the large model with the desired safety property, and also allows the model to remain helpful to benign queries. Furthermore, SSD accelerates the inference time, thanks to the speculative sampling design.
Towards an Account of Complementarities and Context-Dependence
Hong Joo Ryoo
Modern physics proposals present deep tensions between seemingly contradictory descriptions of reality. Views of wave-particle duality, black hole complementarity, and the Unruh effect demand explanations that shift depending on how a system is observed. However, traditional models of scientific explanation impose a fixed structure that fails to account for varying observational contexts. This paper introduces context-dependent mapping, a framework that reorganizes physical laws into self-consistent subsets structured around what can actually be observed in a given context. By doing so, it provides a principled way to integrate complementarity into the philosophy of explanation.
en
physics.hist-ph, quant-ph
Causes in neuron diagrams, and testing causal reasoning in Large Language Models. A glimpse of the future of philosophy?
Louis Vervoort, Vitaly Nikolaev
We propose a test for abstract causal reasoning in AI, based on scholarship in the philosophy of causation, in particular on the neuron diagrams popularized by D. Lewis. We illustrate the test on advanced Large Language Models (ChatGPT, DeepSeek and Gemini). Remarkably, these chatbots are already capable of correctly identifying causes in cases that are hotly debated in the literature. In order to assess the results of these LLMs and future dedicated AI, we propose a definition of cause in neuron diagrams with a wider validity than published hitherto, which challenges the widespread view that such a definition is elusive. We submit that these results are an illustration of how future philosophical research might evolve: as an interplay between human and artificial expertise.
The physicists philosophy of physics
P. J. E. Peebles
I argue that research in physics operates under an implicit community philosophy, and I offer a definition I think physicists would accept, by and large. I compare this definition to what philosophers, sociologists, and historians of science, with physicists, say we are doing.
en
physics.hist-ph, astro-ph.CO
Ernst Haeckel e a controvérsia sobre as imagens de embriões na obra Natürliche Schopfungsgeshichte
Marcelo Viktor Gilge
Ernst Haeckel (1834-1919) foi um renomado pesquisador alemão da segunda metade do século XIX e início do século XX. Parte de sua produção científica foi devotada a defender e divulgar as ideias darwinianas de modificação das espécies em seu país. Entre as ideias de Haeckel, destaca-se a Lei Biogenética Fundamental, na qual ele afirmava que os estágios de desenvolvimento pelos quais passam os embriões recapitulam a história evolutiva do filo. Para explicar essa ideia, na obra Natürliche Schöpfungsgeschichte (História Natural da Criação) de 1868, Haeckel utilizou ilustrações de embriões que foram alvo de críticas e acusações de fraude e plágio. Este artigo tem por objetivos analisar o uso que Ernst Haeckel fez dessas ilustrações, relatando algumas das críticas de cientistas contemporâneas e posteriores e proporcionar um material para atividades pedagógicas voltadas ao ensino de evolução biológica e desenvolvimento embrionário. Em aproximação a análises realizadas por alguns historiadores da ciência, conclui-se que Haeckel se defendeu razoavelmente e que a motivação maior das críticas era o ataque ao darwinismo.
Biology (General), Epistemology. Theory of knowledge
Leveraging Speculative Sampling and KV-Cache Optimizations Together for Generative AI using OpenVINO
Haim Barad, Ekaterina Aidova, Yury Gorbachev
Inference optimizations are critical for improving user experience and reducing infrastructure costs and power consumption. In this article, we illustrate a form of dynamic execution known as speculative sampling to reduce the overall latency of text generation and compare it with standard autoregressive sampling. This can be used together with model-based optimizations (e.g. quantization) to provide an optimized solution. Both sampling methods make use of KV caching. A Jupyter notebook and some sample executions are provided.
Books Received (2022)
C H
Books Received
Speculative philosophy, Ethics
Comparative Analysis of the Diversity of SARS-CoV-2 Lines Circulating in Omsk Region in 2020–2022
E. A. Gradoboeva, Zh. S. Tyulko, A. V. Fadeev
et al.
Relevance. To date, no detailed analysis of the variants of the pathogen circulating at different times on the territory of the Omsk region has been carried out.Aim. Comparative analysis of the diversity of circulating variants of SARSCoV-2 based on molecular genetic data, determine the lines and time of their appearance, compare the data obtained with data from the GISAID database.Materials and methods. Genomewide sequencing of 222 primary and 5 culture (passages on Vero E6 and SPEV cell cultures) samples of SARS-CoV-2 from the Omsk region, collected from April 2020 to February 2022, on Oxford Nanopore Technologies and Illumina platforms, was carried out. Genetic lines were determined in Pangolin. The analysis was performed in MEGA7 and BioEdit.Results. 227 genomewide SARS-CoV-2 sequences were obtained. 222 genomes have been uploaded to the GISAID database. The lines to which the samples belong were determined, phylogenetic trees were constructed for various regions of the SARS-CoV-2 genome, the levels of virus homology were assessed and mutations in the Sprotein region were analyzed.Conclusions. According to the data obtained, it is possible to roughly judge the time of the appearance of a particular variant, its consolidation and distribution in the population, and observe the rare mutations and the circulation of some rare lines. To assess the possibility of significant geographically linked changes in the SARS-CoV-2 in the Omsk region, the data obtained are insufficient. Virus variants circulating in the region are grouped into one cluster with identical variants from other regions or countries. A more pronounced intracluster differentiation of the lines can be observed when analyzing the RBD region. The situation with COVID-19 in the Omsk region generally coincides with that in the whole country and the world. However, this does not exclude the parallel occurrence of certain mutations in remote territories from each other.
Epistemology. Theory of knowledge
Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation
Heming Xia, Tao Ge, Peiyi Wang
et al.
We propose Speculative Decoding (SpecDec), for the first time ever, to formally study exploiting the idea of speculative execution to accelerate autoregressive (AR) decoding. Speculative Decoding has two innovations: Spec-Drafter -- an independent model specially optimized for efficient and accurate drafting -- and Spec-Verification -- a reliable method for verifying the drafted tokens efficiently in the decoding paradigm. Experimental results on various seq2seq tasks including machine translation and abstractive summarization show our approach can achieve around $5\times$ speedup for the popular Transformer architectures with comparable generation quality to beam search decoding, refreshing the impression that the draft-then-verify paradigm introduces only $1.4\times$$\sim$$2\times$ speedup. In addition to the remarkable speedup, we also demonstrate 3 additional advantages of SpecDec, revealing its practical value for accelerating generative models in real-world applications. Our models and codes are available at https://github.com/hemingkx/SpecDec.
COVID-19 Outbreak at Sports Club: Conditions of Occurrence and Causes of the Spread of Infection
A. A. Golubkova, T. A. Platonova, S. S. Smirnova
et al.
Relevance. The new coronavirus infection (COVID-19), which appeared in late 2019 in China, has spread to almost all countries of the world in just a few months. The explosive nature of its spread was accompanied by the formation of large epidemic foci in organizations of various profiles, including leisure and sports. Aims. To establish the conditions and causes of the spread of SARS-CoV-2 among the members of one of the sports clubs based on an in-depth epidemiological analysis. Materials and methods. To study the features of the spread of the SARS-CoV-2 virus in a sports organization, the following documents were used previously developed by the authors and successfully tested in practice: «Act of epidemiological investigation of group and outbreak morbidity of new coronavirus infection (COVID-19) at an enterprise/organization/institution» and «Individual card of a patient with a new coronavirus infection (COVID-19) at the enterprise / organization/institution». In the process of epidemiological investigation, in order to detect SARS-CoV-2 RNA in PCR, a laboratory examination of sports club participants (sick and contact) was conducted, followed by genome-wide sequencing of isolated SARS-CoV-2 viruses on the basis of the Laboratory of Molecular Virology of the A. A. Smorodintsev Influenza Research Institute, which performs these types of studies. Results. Within 17 days, 26 cases of COVID- 19 were registered among the sports team members and staff from the support group (coaching staff, medical staff, administrators), which was 74.3% of their actual number. The majority of patients (76.9%) had mild acute respiratory infection, two (7.7%) had no symptoms, and four (15.4%) had interstitial pneumonia. Of the clinical manifestations of the disease, the most frequent were weakness, fever, headache, muscle and joint pain, difficulty in nasal breathing and serous-mucous discharge from the nose, sore throat, cough, shortness of breath, anosmia and dyspeptic manifestations in the form of diarrhea, nausea or vomiting. The occurrence of the outbreak was the result of the introduction of infection from the opposing team at the tournament. The leading factors that contributed to the spread of COVID-19 among sports club members were the admission to games and training of athletes with acute respiratory infections, prolonged close contact between players during training and competitions, violations in the use of personal protective equipment, compliance with hygiene and hand antiseptics, disinfection measures in the premises of sports institutions and defects in the implementation of the regulations for the examination of teams for SARS-CoV-2 during tournaments. Conclusion. Based on the results of the study, data were obtained on the features of the spread of SARS-CoV-2 in sports organizations, which can be used in conducting preventive and anti-epidemic measures in sports and leisure institutions.
Epistemology. Theory of knowledge
EDITORIAL
Rafael dos Reis Ferreira, João Antonio de Moraes, Pedro Bravo de Souza
et al.
A Kínesis – Revista de Estudos dos Pós-Graduandos em Filosofia apresenta para a comunidade acadêmica filosófica mais uma edição, o Volume 14, Número 36 (2022). Publicamos 23 artigos e 2 traduções. Agradecemos aos pesquisadores que compõem o Conselho Científico da Kínesis e também aos pareceristas ad hoc pela colaboração e disponibilidade permanente para atender nossas solicitações de parecer. Agradecemos, também, aos autores pesquisadores por confiarem a submissão e publicação de suas pesquisas à Kínesis. Convidamos nossos leitores para apreciarem mais este número.
Speculative philosophy, Philosophy (General)
The Quantum Revolution in Philosophy (Book Review)
Eddy Keming Chen
In this thought-provoking book, Richard Healey proposes a new interpretation of quantum theory inspired by pragmatist philosophy. Healey puts forward the interpretation as an alternative to realist quantum theories on the one hand such as Bohmian mechanics, spontaneous collapse theories, and many-worlds interpretations, which are different proposals for describing what the quantum world is like and what the basic laws of physics are, and non-realist interpretations on the other hand such as quantum Bayesianism, which proposes to understand quantum theory as describing agents' subjective epistemic states. The central idea of Healey's proposal is to understand quantum theory as providing not a description of the physical world but a set of authoritative and objectively correct prescriptions about how agents should act. The book provides a detailed development and defense of that idea, and it contains interesting discussions about a wide range of philosophical issues such as representation, probability, explanation, causation, objectivity, meaning, and fundamentality. Healey's project is at the intersection of physics and philosophy. The book is divided into two parts. Part I of the book discusses the foundational questions in quantum theory from the perspective of the prescriptive interpretation. In Part II, Healey discusses the philosophical implications of the view. Both parts are written in a way that is largely accessible to non-specialists. In this brief book review, I will focus on two questions: (1) How does Healey's idea work? (2) What reasons are there to believe in it?
en
physics.hist-ph, quant-ph
Prevalence of Cardiovascular Disease Risk Factors in Vologda Oblast Districts
N. Kh. Svanadze, R. A. Kasimov, A. A. Orlovsky
et al.
Relevance. There are large regional disparities in prevalence of non-communicable disease risk factors, as well as in the cardiovascular disease (CVD) incidence and mortality rates in Russian Federation (RF). Aim. To demonstrate the disparities in prevalence of CVD risk factors between Vologda Oblast districts. Materials and methods. Databases created in 2009 at the State-financed health institution of the Vologda Oblast «Vologda Regional Center for Medical Prevention», based on the results of a survey conducted within the framework of the World Health Organization CINDI program. CINDI questionnaire; cross-sectional study; the data was processed using R programming language and the Statistica software package 12. Results. The most common behavioral CVD risk factors in different Vologda Oblast districts included inadequate fruits and vegetables consumption (30–90%) and alcohol abuse (40–80%); hypertension (40–60%), overweight and obesity (30–55%) were the most frequent biological CVD risk factors; the most prevalent socio-economic risk factors included low education level (75–90%) and unemployment (20–40%). Participants residing in rural municipalities differed from urban okrugs (cities) dwellers in a higher prevalence of smoking (p < 0.01), alcohol abuse (p < 0.001), inadequate fruits and vegetables consumption (p < 0.0001), overweight and obesity (p < 0.05), unemployment (p < 0.0001), low education level (p < 0.0001), as well as a low overall assessment of their health (p < 0.05). Conclusions. We detected disparities in CVD risk factors prevalence between Vologda Oblast districts in 2009. Both behavioral and biological CVD risk factors were more common in participants from rural municipalities. The CVD risk factors distribution between the RF subjects’ districts requires further scientific research.
Epistemology. Theory of knowledge
The Notion of Power in Hans Jonas’ Das Prinzip Verantwortung (The Imperative of Responsibility)
Piotr Rosół
The questions concerning control over the environment are becoming increasingly more significant. From ecology to medicine, from bioethics to transhumanism, there are many different issues reflected and acted upon, which have an important common element, namely underlying premises about the relationship between the natural biosphere and humans. Hans Jonas, in his book The Imperative of Responsibility: In Search for an Ethics for the Technological Age, first published over 40 years ago in German as Das Prinzip Verantwortung: Versuch einer Ethik für die technologische Zivilisation, proposed philosophical and ethical foundations for understanding this relationship. One of the underestimated Hans Jonas’ reflections concerns the characteristic of power. The paper analyzes the concept of power presented along with the imperative of responsibility, as well as three different degrees of power characteristic for Jonas’ approach to the notion of power. The paper presents this structure of power and the underlying philosophical considerations that constitute a basis for it. It also provides an argument that these considerations may be used as a framework for understanding many contemporary challenges and ethical responsibility in the technological age.
Speculative philosophy, Philosophy (General)
QATIPANA: Processes of Individuation on the Relationship Between Art, Machine and Natural Systems
Renzo Ch. Filinich Orozco, Tamara J. Chibey Rivas
The present research turns around the concepts and processes of Becoming and Individuation where it evidences a functional model based on the articulation of an informational processing system based on the approaches of the philosopher Gilbert Simondon. It aims to model a sensorimotor cycle performed by the cognitive system of an Artificial Intelligence agent. To establish this model of biological inspiration, we use the concepts of information in cybernetics by Norbert Wiener, information and modulation in Gilbert Simondon and the notion of machine performativity in light of Bernard Stiegler's ideas. Although the architecture that we have called Qatipana (Quechua word that denotes the flow of information processing systems) cannot be considered as a systems theory, it has the utility of being able to explain some empirical observations that we also present here. In conclusion, the implications and limitations of this model and the research that is being carried out to present its utility and probability as a model of the algorithmic cognitive system are part of the questions of communication and affect in the decisions provided by the automatic system.
Speculative philosophy, Ethics
Physics Needs Philosophy. Philosophy Needs Physics
Carlo Rovelli
Contrary to claims about the irrelevance of philosophy for science, I argue that philosophy has had, and still has, far more influence on physics than is commonly assumed. I maintain that the current anti-philosophical ideology has had damaging effects on the fertility of science. I also suggest that recent important empirical results, such as the detection of the Higgs particle and gravitational waves, and the failure to detect supersymmetry where many expected to find it, question the validity of certain philosophical assumptions common among theoretical physicists, inviting us to engage in a clearer philosophical reflection on scientific method.
Book Review: Byung-Chul Han, La società della stanchezza, nottetempo, Milano 2012
Salottolo, Delio
Epistemology. Theory of knowledge, Ethics