Stephane Requena

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a… Voir plus

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). We find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, we publicly release our models and code under the Responsible AI License. Voir moins

Over the past four years, the Big Data and Exascale Computing (BDEC) project organized a series of five international workshops that aimed to explore the ways in which the new forms of data-centric discovery introduced by the ongoing revolution in high-end data analysis (HDA) might be integrated with the established, simulation-centric paradigm of the high-performance computing (HPC) community. Based on those meetings, we argue that the rapid proliferation of digital data generators, the… Voir plus

Over the past four years, the Big Data and Exascale Computing (BDEC) project organized a series of five international workshops that aimed to explore the ways in which the new forms of data-centric discovery introduced by the ongoing revolution in high-end data analysis (HDA) might be integrated with the established, simulation-centric paradigm of the high-performance computing (HPC) community. Based on those meetings, we argue that the rapid proliferation of digital data generators, the unprecedented growth in the volume and diversity of the data they generate, and the intense evolution of the methods for analyzing and using that data are radically reshaping the landscape of scientific computing. The most critical problems involve the logistics of wide-area, multistage workflows that will move back and forth across the computing continuum, between the multitude of distributed sensors, instruments and other devices at the networks edge, and the centralized resources of commercial clouds and HPC centers. Voir moins

"From telescopes to microscopes, from vacuums to hyperbaric chambers, from sonar waves to laser beams, scientists have perpetually strived to apply technology and invention to new frontiers of scientific advancement. Along the way, they have used abacuses, slide rules, calculators, computers, and - today - supercomputers, to crunch the increasingly complicated calculations used to understand and predict natural phenomena. In the course of practicality, science begets engineering, and… Voir plus

"From telescopes to microscopes, from vacuums to hyperbaric chambers, from sonar waves to laser beams, scientists have perpetually strived to apply technology and invention to new frontiers of scientific advancement. Along the way, they have used abacuses, slide rules, calculators, computers, and - today - supercomputers, to crunch the increasingly complicated calculations used to understand and predict natural phenomena. In the course of practicality, science begets engineering, and engineering begets innovation. Science, and therefore supercomputing, finds its way into industry around the world, affecting our lives in ways that are simple (cleaner clothes, faster speedboats, and more spin on a seven-iron) and profound (fewer drug interactions, safer cars, and new energy sources). Altogether, industry consumes more than half of all high performance computing usage worldwide. This book tells that story."

Industrial Applications of High-Performance Computing: Best Global Practices supplies computer engineers and researchers with a state-of-the-art supercomputing reference. This book also keeps policymakers and industrial decision-makers informed about the economic impact of these powerful technological investments. Voir moins

We have performed the first-ever numerical N-body simulations of the full observable universe with Dark Energy (DEUS "Dark Energy Universe Simulation" FUR "Full Universe Run"). Three different Dark Energy models have been performed. These have evolved 550 billion particles on an Adaptive Mesh Refinement grid with more than two and half trillion computing points along the entire evolutionary history of the universe and across 6 order of magnitudes length scales, from the size of the Milky Way to… Voir plus

We have performed the first-ever numerical N-body simulations of the full observable universe with Dark Energy (DEUS "Dark Energy Universe Simulation" FUR "Full Universe Run"). Three different Dark Energy models have been performed. These have evolved 550 billion particles on an Adaptive Mesh Refinement grid with more than two and half trillion computing points along the entire evolutionary history of the universe and across 6 order of magnitudes length scales, from the size of the Milky Way to that of the whole observable Universe. To date, these are the largest and most advanced complete cosmological simulations ever run. It provides unique information on the formation and evolution of the largest structure in the universe and an exceptional support to future observational programs dedicated to mapping the distribution of matter and galaxies in the universe. The simulations have run on 4752 (of 5040) thin nodes of BULL’ supercomputer CURIE, using more than 300 TB of memory for 10 million hours of computing time each. About 150 PBytes of data were generated throughout the run. Using an advanced and innovative reduction workflow the amount of useful stored data has been reduced to 1.5 PBytes. Voir moins

We performed a massive N-body simulation of the full observable universe. This has evolved 550 billion particles on an Adaptive Mesh Refinement grid with more than two trillion computing points along the entire evolutionary history of the Universe, and across 6 orders of magnitudes length scales, from the size of the Milky Way to the whole observable Universe. To date, this is the largest and most advanced cosmological simulation ever run. It will have a major scientific impact and provide an… Voir plus

We performed a massive N-body simulation of the full observable universe. This has evolved 550 billion particles on an Adaptive Mesh Refinement grid with more than two trillion computing points along the entire evolutionary history of the Universe, and across 6 orders of magnitudes length scales, from the size of the Milky Way to the whole observable Universe. To date, this is the largest and most advanced cosmological simulation ever run. It will have a major scientific impact and provide an exceptional support to future observational programs dedicated to mapping the distribution of matter and galaxies in the Universe. The simulation has run on 4752 (of 5040) thin nodes of BULL supercomputer CURIE, using 300 TB of memory for 10 million hours of computing time. 50 PBytes of rough data were generated throughout the run, reduced to a useful amount of 500 TBytes using an advanced and innovative reduction workflow. Voir moins

The success of the gaming industry is now pushing processor technology like we have never seen before. Since recent graphics processors (GPU’s) have been improving both their programmability as well as have been adding more and more
floating point processing, it makes them very appealing as accelerators for general purpose computing. This minisymposium gives an overview of some of these advancements by bringing together experts working on the development of techniques
and tools that… Voir plus

The success of the gaming industry is now pushing processor technology like we have never seen before. Since recent graphics processors (GPU’s) have been improving both their programmability as well as have been adding more and more
floating point processing, it makes them very appealing as accelerators for general purpose computing. This minisymposium gives an overview of some of these advancements by bringing together experts working on the development of techniques
and tools that improve the programmability of GPU’s as well as the experts interested in utilizing the computational power of GPU’ scientific applications. This first EuroGPU Minisymposium brought together severl experts working on the development
of techniques and tools that improve the programmability of GPU’s as well as the experts interested in utilizing the computational power of GPU’s for scientific applications. This short summary thus gives a very useful, but quick overview of
some of the major recent advancement in modern GPU computing. Voir moins

Our goal was to develop an hybrid version running on GPUs using CUDA of our internal linear solver used for reservoir and bassin modeling. this solver use a an ilu(0) or an AMG preconditioner and a BiCGStab solver as backend.

In recent years the 3D modelling and simulation of sedimentary basins has become an integral part of the exploration of reservoirs for almost all major oil companies. The most computing intensive part of the simulation consists in solving numerically a complex system of non linear partial differential equations predicting the entire physical and chemical processes of hydrocarbon generation and accumulation through geological times. With a price/performance ratio and with services in constant… Voir plus

In recent years the 3D modelling and simulation of sedimentary basins has become an integral part of the exploration of reservoirs for almost all major oil companies. The most computing intensive part of the simulation consists in solving numerically a complex system of non linear partial differential equations predicting the entire physical and chemical processes of hydrocarbon generation and accumulation through geological times. With a price/performance ratio and with services in constant progress Linux clusters have become a very attractive alternative to “traditional” high performance computing facilities. ...
Voir moins

The simulation of sedimentary basins aims at reconstructing its historical evolution in order to provide quantitative predictions about phenomena leading to hydrocarbon accumulations. The kernel of this simulation is the numerical solution of a complex system of non-linear partial differential equations (PDE) of mixed parabolic-hyperbolic type in 3D. A discretisation and linearisation of this system leads to very large, ill-conditioned, non-symmetric systems of linear equations with three… Voir plus

The simulation of sedimentary basins aims at reconstructing its historical evolution in order to provide quantitative predictions about phenomena leading to hydrocarbon accumulations. The kernel of this simulation is the numerical solution of a complex system of non-linear partial differential equations (PDE) of mixed parabolic-hyperbolic type in 3D. A discretisation and linearisation of this system leads to very large, ill-conditioned, non-symmetric systems of linear equations with three unknowns per mesh cell, i.e. pressure, geostatic load, and oil saturation.
This article describes the parallel version of a preconditioner for these systems, presented in its sequential form in [7]. It consists of three steps: in the first step a local decoupling of the pressure and saturation unknowns aims at concentrating in the “pressure block” the elliptic part of the system which is then, in the second step, preconditioned by AMG. The third step finally consists in recoupling the equations. Each step is efficiently parallelised using a partitioning of the domain into vertical layers along the y-axis and a distributed memory model within the PETSc library (Argonne National Laboratory, IL). The main new ingredient in the parallel version is a parallel AMG preconditioner for the pressure block, for which we use the BoomerAMG implementation in the hypre library [4].
Numerical results on real case studies, exhibit (i) a significant reduction of CPU times, up to a factor 5 with respect to a block Jacobi preconditioner with an ILU(0) factorisation of each block, (ii) robustness with respect to heterogeneities, anisotropies and high migration ratios, and (iii) a speedup of up to 4 on 8 processors. Voir moins