Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “code optimization”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 847 records · Page 47

Neutral beams for the burning plasma era: Simulations and modeling of deuterium ion beams for plasma heating (CRADA Final Report)

Together with Realta Fusion, we set up several simulation scenarios and ran simulations using the WarpX code which is developed by LBNL and especially by members of the LBNL team. We investigated beam transport and beam neutralization and ran optimization and benchmark simulations. To achieve this, several new features were added to the warpX code. These code additions will also be beneficial to other users of warpX including several fusion companies and researchers at National Laboratories. Furthermore, we supported Realta Fusion in setting up their own instances of WarpX. We presented our results at the American Physical Society Division of Plasma Physics meeting and are in the process of writing a publication that will be submitted to a peer-reviewed journal to share our results with the broader community.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

The Modeling of Advanced BWR Fuel Designs with the NRC Fuel Depletion Codes PARCS/PATHS

The PATHS (PARCS Advanced Thermal Hydraulic Solver) code was developed at the University of Michigan in support of U.S. Nuclear Regulatory Commission research to solve the steady-state, two-phase, thermal-hydraulic equations for a boiling water reactor (BWR) and to provide thermal-hydraulic feedback for BWR depletion calculations with the neutronics code PARCS (Purdue Advanced Reactor Core Simulator). The simplified solution methodology, including a three-equation drift flux formulation and an optimized iteration scheme, yields very fast run times in comparison to conventional thermal-hydraulic systems codes used in the industry, while still retaining sufficient accuracy for applications such as BWR depletion calculations. Lastly, the capability to model advanced BWR fuel designs with part-length fuel rods and heterogeneous axial channel flow geometry has been implemented in PATHS, and the code has been validated against previously benchmarked advanced core simulators as well as BWR plant and experimental data. We describe the modifications to the codes and the results of the validation in this paper.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Towards performance portability in the Spark astrophysical magnetohydrodynamics solver in the Flash-X simulation framework

Simulations of core-collapse supernovae, and other astrophysical phenomena, are quintessential extreme-scale computing challenges. For core-collapse supernova simulations to be carried out by the ExaStar project under the Exascale Computing Project umbrella, a robust, efficient, and state-of-the-art magnetohydrodynamics solver is a critical requirement. In Flash-X, the primary software instrument for ExaStar, a new magnetohydrodynamics solver has been designed and implemented from the ground up to achieve accuracy and efficiency for simulations of complex astrophysical flows. This new solver, dubbed Spark, uses high-order spatial reconstruction, Runge-Kutta time integration, and an efficient cell-centered approach to satisfying the divergence-free condition for the magnetic fields. Spark was written to be optimized for data locality in cache hierarchy of CPUs. Since data locality optimizations for cache hierarchy are not directly compatible with those of accelerators, we have taken the approach of using program synthesis to avoid massive amounts of code replication that would be necessary if we were to maintain two different versions of the solver. Our program synthesis relies on a simple key-dictionary approach, implemented in python, that enables us to assemble the version of the solver suitable for the target hardware from code fragments identified by specific keys. In this work, we describe the data locality optimizations of the solver for CPUs and accelerators and the program synthesis tools that enable this portability. We also detail the parallel performance of Spark for both CPUs and accelerators.

97 MATHEMATICS AND COMPUTING↗

Distributed Energy Resource Optimization Using a Software as Service (SaaS) Approach at the University of California, Davis Campus

Together with OSIsoft LLC as its private sector partner and matching sponsor, the Lawrence Berkeley National Laboratory (Berkeley Lab) won an FY09 Technology Commercialization Fund (TCF) grant from the U.S. Department of Energy. The goal of the project is to commercialize Berkeley Lab's optimizing program, the Distributed Energy Resources Customer Adoption Model (DER-CAM) using a software as a service (SaaS) model with OSIsoft as its first non-scientific user. OSIsoft could in turn provide optimization capability to its software clients. In this way, energy efficiency and/or carbon minimizing strategies could be made readily available to commercial and industrial facilities. Specialized versions of DER-CAM dedicated to solving OSIsoft's customer problems have been set up on a server at Berkeley Lab. The objective of DER-CAM is to minimize the cost of technology adoption and operation or carbon emissions, or combinations thereof. DER-CAM determines which technologies should be installed and operated based on specific site load, price information, and performance data for available equipment options. An established user of OSIsoft's PI software suite, the University of California, Davis (UCD), was selected as a demonstration site for this project. UCD's participation in the project is driven by its motivation to reduce its carbon emissions. The campus currently buys electricity economically through the Western Area Power Administration (WAPA). The campus does not therefore face compelling cost incentives to improve the efficiency of its operations, but is nonetheless motivated to lower the carbon footprint of its buildings. Berkeley Lab attempted to demonstrate a scenario wherein UCD is forced to purchase electricity on a standard time-of-use tariff from Pacific Gas and Electric (PG&E), which is a concern to Facilities staff. Additionally, DER-CAM has been set up to consider the variability of carbon emissions throughout the day and seasons. Two distinct analyses of value to UCD are possible using this approach. First, optimal investment choices for buildings under the two alternative objectives can be derived. Second, a week-ahead building operations forecaster has been written that executes DER-CAM to find an optimal operating schedule for buildings given their expected building energy services requirements, electricity prices, and local weather. As part of its matching contribution, OSIsoft provided a full implementation of PI and a server to install it on at Berkeley Lab. Using the PItoPI protocol, this gives Berkeley Lab researchers direct access to UCD's PI data base. However, this arrangement is in itself inadequate for performing optimizations. Additional data not included in UCD's PI database would be needed and the campus was not able to provide this information. This report details the process, results, and lessons learned of this commercialization project.

25 ENERGY STORAGE↗

Using Forward-Reversal Integration to Optimize EIC's e-ring Dynamic Aperture

The forward-reversal integration has been proved efficient in optimizing the dynamic aperture for light source rings, which usually have a multi-fold periodical lattice structure. This method has been integrated into the ELEGANT code since the version 2019.4. In this technote, this approach has been used to optimize the dynamic aperture of a collider ring, i.e., the future electron ion collider’s (EIC) e-ring.

43 PARTICLE ACCELERATORS↗

NNSA ASC Exascale Environment Planning, Applications Working Group, Report February 2011

The scope of the Apps WG covers three areas of interest: Physics and Engineering Models (PEM), multi-physics Integrated Codes (IC), and Verification and Validation (V&V). Each places different demands on the exascale environment. The exascale challenge will be to provide environments that optimize all three. PEM serve as a test bed for both model development and 'best practices' for IC code development, as well as their use as standalone codes to improve scientific understanding. Rapidly achieving reasonable performance for a small team is the key to maintaining PEM innovation. Thus, the environment must provide the ability to develop portable code at a higher level of abstraction, which can then be tuned, as needed. PEM concentrate their computational footprint in one or a few kernels that must perform efficiently. Their comparative simplicity permits extreme optimization, so the environment must provide the ability to exercise significant control over the lower software and hardware levels. IC serve as the underlying software tools employed for most ASC problems of interest. Often coupling dozens of physics models into very large, very complex applications, ICs are usually the product of hundreds of staff-years of development, with lifetimes measured in decades. Thus, emphasis is placed on portability, maintainability and overall performance, with optimization done on the whole rather than on individual parts. The exascale environment must provide a high-level standardized programming model with effective tools and mechanisms for fault detection and remediation. Finally, V&V addresses the infrastructure and methods to facilitate the assessment of code and model suitability for applications, and uncertainty quantification (UQ) methods for assessment and quantification of margins of uncertainty (QMU). V&V employs both PEM and IC, with somewhat differing goals, i.e., parameter studies and error assessments to determine both the quality of the calculation and to estimate expected deviations of simulations from experiments. The exascale environment must provide a performance envelope suitable both for capacity calculations (high through-put) and full system capability runs (high performance). Analysis of the results place shared demand on both the I/O as well as the visualization subsystems.

97 MATHEMATICS AND COMPUTING↗

DaCe - Data Centric Parallel Programming

DaCe is a parallel programming framework that takes code in Python/NumPy and other programming languages, and maps it to high-performance CPU, GPU, and FPGA programs, which can be optimized to achieve state-of-the-art. Internally, DaCe uses the Stateful DataFlow multiGraph (SDFG) data-centric intermediate representation: A transformable, interactive representation of code based on data movement. Since the input code and the SDFG are separate, it is possible to optimize a program without changing its source, so that it stays readable. On the other hand, transformations are customizable and user-extensible, so they can be written once and reused in many applications. With data-centric parallel programming, we enable direct knowledge transfer of performance optimization, regardless of the application or the target processor.

Ben-Nun, Tal↗

Introduction to an MCGIDI Mini-App and Performance Comparisons with XSBench

In high-performance Monte Carlo (MC) radiation transport codes, cross section lookups often account for the majority of computational expens. In response to this observation, a number of nuclear data mini-apps have been developed to profile and optimize the lookup process. A review of these mini-apps can be found in Ref. In contrast to a fully fledged MC code, nuclear data mini-apps do not feature particle tracking; instead, the cross section lookup process is considered in isolation, a simplification that allows for easier implementation and evaluation of various lookup schemes and techniques. GIDI+, as described on its public repository1, is ...a collection of C++ libraries for accessing evaluated and processed nuclear data stored in the Generalized Nuclear Database Structure (GNDS). In addition to reading GNDS files, GIDI+ has functions to sum and collapse multi-group data as needed by deterministic transport codes, and to sample GNDS data as needed by Monte Carlo transport codes. Certain modes of Mercury, a LLNL-developed MC radiation transport code, utilize GIDI+ for cross section lookups. Specifically, Mercury uses the MCGIDI library contained within GIDI+ to perform lookups on GPUs. A mini-app exercising MCGIDI’s cross section lookup capabilities would prove useful for optimization, which in turn could improve the performance of client codes such as Mercury. Such a mini-app could also be used to assess the performance of MCGIDI compared to other nuclear data mini-apps. In this document, we introduce a nuclear data mini-app built with the MCGIDI library. The MCGIDI mini-app (MCGIDI-MA) was developed to closely resemble the methodology present in XSBench, a nuclear data mini-app developed at Argonne National Laboratory. The remainder of this section provides an overview on XSBench. Section 2 describes MCGIDI-MA and its usage. Section 3 contains performance results of MCGIDI-MA on LLNL’s Lassen and Quartz compute platforms, as well as comparisons to XSBench performance where appropriate. In Section 4, we present a summary of our findings.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Computational modeling of Krypton gas puffs with tailored mass density profiles on Z

Large diameter multi-shell gas puffs rapidly imploded by high current (~20 MA, ~100 ns) on the Z generator of Sandia National Laboratories are able to produce high-intensity Krypton K-shell emission at ~13 keV. Efficiently radiating at these high photon energies is a significant challenge which requires the careful design and optimization of the gas distribution. To facilitate this, we hydrodynamically model the gas flow out of the nozzle and then model its implosion using a 3-dimensional resistive, radiative MHD code (GORGON). This approach enables us to iterate between modeling the implosion and gas flow from the nozzle to optimize radiative output from this combined system. Furthermore, guided by our implosion calculations, we have designed gas profiles that help mitigate disruption from Magneto-Rayleigh–Taylor implosion instabilities, while preserving sufficient kinetic energy to thermalize to the high temperatures required for K-shell emission.

97 MATHEMATICS AND COMPUTING↗

Sparse coding of pathology slides compared to transfer learning with deep neural networks

Background Histopathology images of tumor biopsies present unique challenges for applying machine learning to the diagnosis and treatment of cancer. The pathology slides are high resolution, often exceeding 1GB, have non-uniform dimensions, and often contain multiple tissue slices of varying sizes surrounded by large empty regions. The locations of abnormal or cancerous cells, which may constitute a small portion of any given tissue sample, are not annotated. Cancer image datasets are also extremely imbalanced, with most slides being associated with relatively common cancers. Since deep representations trained on natural photographs are unlikely to be optimal for classifying pathology slide images, which have different spectral ranges and spatial structure, we here describe an approach for learning features and inferring representations of cancer pathology slides based on sparse coding. Results We show that conventional transfer learning using a state-of-the-art deep learning architecture pre-trained on ImageNet (RESNET) and fine tuned for a binary tumor/no-tumor classification task achieved between 85% and 86% accuracy. However, when all layers up to the last convolutional layer in RESNET are replaced with a single feature map inferred via a sparse coding using a dictionary optimized for sparse reconstruction of unlabeled pathology slides, classification performance improves to over 93%, corresponding to a 54% error reduction. Conclusions We conclude that a feature dictionary optimized for biomedical imagery may in general support better classification performance than does conventional transfer learning using a dictionary pre-trained on natural images.

59 BASIC BIOLOGICAL SCIENCES↗

A novel software and conceptual design of the hardware platform for intensity modulated radiation therapy

Purpose: To deliver high quality intensity modulated radiotherapy (IMRT) using a novel generalized sparse orthogonal collimators (SOCs), the authors introduce a novel direct aperture optimization (DAO) approach based on discrete rectangular representation. Methods: A total of seven patients—two glioblastoma multiforme, three head & neck (including one with three prescription doses), and two lung—were included. 20 noncoplanar beams were selected using a column generation and pricing optimization method. The SOC is a generalized conventional orthogonal collimators with N leaves in each collimator bank, where N = 1, 2, or 4. SOC degenerates to conventional jaws when N = 1. For SOC-based IMRT, rectangular aperture optimization (RAO) was performed to optimize the fluence maps using rectangular representation, producing fluence maps that can be directly converted into a set of deliverable rectangular apertures. In order to optimize the dose distribution and minimize the number of apertures used, the overall objective was formulated to incorporate an L2 penalty reflecting the difference between the prescription and the projected doses, and an L1 sparsity regularization term to encourage a low number of nonzero rectangular basis coefficients. The optimization problem was solved using the Chambolle–Pock algorithm, a first-order primal–dual algorithm. Performance of RAO was compared to conventional two-step IMRT optimization including fluence map optimization and direct stratification for multileaf collimator (MLC) segmentation (DMS) using the same number of segments. For the RAO plans, segment travel time for SOC delivery was evaluated for the N = 1, N = 2, and N = 4 SOC designs to characterize the improvement in delivery efficiency as a function of N. Results: Comparable PTV dose homogeneity and coverage were observed between the RAO and the DMS plans. The RAO plans were slightly superior to the DMS plans in sparing critical structures. On average, the maximum and mean critical organ doses were reduced by 1.94% and 1.44% of the prescription dose. The average number of delivery segments was 12.68 segments per beam for both the RAO and DMS plans. The N = 2 and N = 4 SOC designs were, on average, 1.56 and 1.80 times more efficient than the N = 1 SOC design to deliver. The mean aperture size produced by the RAO plans was 3.9 times larger than that of the DMS plans. Conclusions: The DAO and dose domain optimization approach enabled high quality IMRT plans using a low-complexity collimator setup. The dosimetric quality is comparable or slightly superior to conventional MLC-based IMRT plans using the same number of delivery segments. The SOC IMRT delivery efficiency can be significantly improved by increasing the leaf numbers, but the number is still significantly lower than the number of leaves in a typical MLC.

61 RADIATION PROTECTION AND DOSIMETRY↗

RAVEN Beta Release

This documents the release of the Risk Analysis Virtual Environment (RAVEN) code. A description of the RAVEN code is provided, and discussion of the release process for the M2LW-16IN0704045 milestone. The RAVEN code is a generic software framework to perform parametric and probabilistic analysis based on the response of complex system codes. RAVEN is capable of investigating the system response as well as the input space using Monte Carlo, Grid, or Latin Hyper Cube sampling schemes, but its strength is focused toward system feature discovery, such as limit surfaces, separating regions of the input space leading to system failure, using dynamic supervised learning techniques. RAVEN has now increased in maturity enough for the Beta 1.0 release.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

SU-D-204-01: A Methodology Based On Machine Learning and Quantum Clustering to Predict Lung SBRT Dosimetric Endpoints From Patient Specific Anatomic Features

Purpose: To develop a data-mining methodology based on quantum clustering and machine learning to predict expected dosimetric endpoints for lung SBRT applications based on patient-specific anatomic features. Methods: Ninety-three patients who received lung SBRT at our clinic from 2011–2013 were retrospectively identified. Planning information was acquired for each patient, from which various features were extracted using in-house semi-automatic software. Anatomic features included tumor-to-OAR distances, tumor location, total-lung-volume, GTV and ITV. Dosimetric endpoints were adopted from RTOG-0195 recommendations, and consisted of various OAR-specific partial-volume doses and maximum point-doses. First, PCA analysis and unsupervised quantum-clustering was used to explore the feature-space to identify potentially strong classifiers. Secondly, a multi-class logistic regression algorithm was developed and trained to predict dose-volume endpoints based on patient-specific anatomic features. Classes were defined by discretizing the dose-volume data, and the feature-space was zero-mean normalized. Fitting parameters were determined by minimizing a regularized cost function, and optimization was performed via gradient descent. As a pilot study, the model was tested on two esophageal dosimetric planning endpoints (maximum point-dose, dose-to-5cc), and its generalizability was evaluated with leave-one-out cross-validation. Results: Quantum-Clustering demonstrated a strong separation of feature-space at 15Gy across the first-and-second Principle Components of the data when the dosimetric endpoints were retrospectively identified. Maximum point dose prediction to the esophagus demonstrated a cross-validation accuracy of 87%, and the maximum dose to 5cc demonstrated a respective value of 79%. The largest optimized weighting factor was placed on GTV-to-esophagus distance (a factor of 10 greater than the second largest weighting factor), indicating an intuitively strong correlation between this feature and both endpoints. Conclusion: This pilot study shows that it is feasible to predict dose-volume endpoints based on patient-specific anatomic features. The developed methodology can potentially help to identify patients at risk for higher OAR doses, thus improving the efficiency of treatment planning. R01-184173.

60 APPLIED LIFE SCIENCES↗

Optimizing the emission, propagation, and focusing of an intense electron beam

Intense electron beams can be used to study the dynamical response of materials under shocks in order to adjust the models developed for hydrodynamics simulations. We present in this paper a characterization of beams produced in a field emission diode coupled to the generator RKA at CEA/CESTA. Cherenkov emission, produced by the beam interacting in a fused silica disk, was observed by fast optical cameras to estimate beam homogeneity. GEANT4 simulations were performed to estimate the transfer function of the silica target and to optimize the anode foil. First, we chose the best cathode material available among the most common materials used in field emission systems. In addition, we found that by optimization of the anode thickness, we could improve the spatial homogeneity of the beam which is of prime importance for computing the interaction of the beam with materials. Next, we changed the beam fluence by increasing the beam current and by reducing the beam radius. Finally, we studied the propagation and focusing of the electron beam in low pressure gases and observed that we could use self-magnetic field focusing in order to increase beam fluence at the target location. The experimental results are in good agreement with PIC simulations.

43 PARTICLE ACCELERATORS↗

PHOENIX Electrostatic Design

PHOENIX (Portable, High-efficiency, Optimal ENergy Imaging X-rays) is a quasi-DC, electrostatic, vacuum-diode designed as a portable x-ray source with national defense and commercial applications. The patent-pending PHOENIX concept combines a megavoltage, Cockroft-Walton voltage multiplier with a Van de-Graaff electrostatic charge-storage dome to create a vacuum-diode suitable for x-ray production. Naturally this structure must minimize internal electric fields to reduce electrical breakdown while simultaneously reducing size and weight to enhance portability. In this paper we describe the optimization process and model results obtained using the COMSOL multi-physics code. We describe three models: a prototype model built to demonstrate the PHOENIX concept as part of Laboratory Directed Research and Development (LDRD) Mission Foundation Research (MFR) Phase-I , a “back-of-the-envelope” design used as a starting point for further COMSOL optimization, and finally, the optimized geometry implemented in the MFR Phase-II. In all cases compromises resulting from cost, schedule, and manufacturing constraints were taken into account as the design progressed.

42 ENGINEERING↗

SU-D-18A-02: Towards Real-Time On-Board Volumetric Image Reconstruction for Intrafraction Target Verification in Radiation Therapy

Purpose: To expedite on-board volumetric image reconstruction from limited-angle kV—MV projections for intrafraction verification. Methods: A limited-angle intrafraction verification (LIVE) system has recently been developed for real-time volumetric verification of moving targets, using limited-angle kV—MV projections. Currently, it is challenged by the intensive computational load of the prior-knowledge-based reconstruction method. To accelerate LIVE, we restructure the software pipeline to make it adaptable to model and algorithm parameter changes, while enabling efficient utilization of rapidly advancing, modern computer architectures. In particular, an innovative two-level parallelization scheme has been designed: At the macroscopic level, data and operations are adaptively partitioned, taking into account algorithmic parameters and the processing capacity or constraints of underlying hardware. The control and data flows of the pipeline are scheduled in such a way as to maximize operation concurrency and minimize total processing time. At the microscopic level, the partitioned functions act as independent modules, operating on data partitions in parallel. Each module is pre-parallelized and optimized for multi-core processors (CPUs) and graphics processing units (GPUs). Results: We present results from a parallel prototype, where most of the controls and module parallelization are carried out via Matlab and its Parallel Computing Toolbox. The reconstruction is 5 times faster on a data-set of twice the size, compared to recently reported results, without compromising on algorithmic optimization control. Conclusion: The prototype implementation and its results have served to assess the efficacy of our system concept. While a production implementation will yield much higher processing rates by approaching full-capacity utilization of CPUs and GPUs, some mutual constraints between algorithmic flow and architecture specifics remain. Based on a careful analysis of the prototype performance, it will be feasible to resolve such issues through appropriate algorithmic modifications or special-purpose hardware, thus enabling target verification in seconds with the LIVE system. This work was partially supported by a research grant from Varian Medical Systems.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

DOE SBIR Phase-1 Report on Hybrid CPU-GPU Parallel Development of the Eulerian-Lagrangian Barracuda Multiphase Program

This report gives the result from the Phase-1 work on demonstrating greater than 10x speedup of the Barracuda computer program using parallel methods and GPU processors (General-Purpose Graphics Processing Unit or Graphics Processing Unit). Phase-1 demonstrated a 12x speedup on a typical Barracuda function using the GPU processor. The problem test case used about 5 million particles and 250,000 Eulerian grid cells. The relative speedup, compared to a single CPU, increases with increased number of particles giving greater than 12x speedup. Phase-1 work provided a path for reformatting data structure modifications to give good parallel performance while keeping a friendly environment for new physics development and code maintenance. The implementation of data structure changes will be in Phase-2. Phase-1 laid the ground work for the complete parallelization of Barracuda in Phase-2, with the caveat that implemented computer practices for parallel programming done in Phase-1 gives immediate speedup in the current Barracuda serial running code. The Phase-1 tasks were completed successfully laying the frame work for Phase-2. The detailed results of Phase-1 are within this document. In general, the speedup of one function would be expected to be higher than the speedup of the entire code because of I/O functions and communication between the algorithms. However, because one of the most difficult Barracuda algorithms was parallelized in Phase-1 and because advanced parallelization methods and proposed parallelization optimization techniques identified in Phase-1 will be used in Phase-2, an overall Barracuda code speedup (relative to a single CPU) is expected to be greater than 10x. This means that a job which takes 30 days to complete will be done in 3 days. Tasks completed in Phase-1 are: Task 1: Profile the entire Barracuda code and select which subroutines are to be parallelized (See Section Choosing a Function to Accelerate) Task 2: Select a GPU consultant company and jointly parallelize subroutines (CPFD chose the small business EMPhotonics for the Phase-1 the technical partner. See Section Technical Objective and Approach) Task 3: Integrate parallel subroutines into Barracuda (See Section Results from Phase-1 and its subsections) Task 4: Testing, refinement, and optimization of parallel methodology (See Section Results from Phase-1 and Section Result Comparison Program) Task 5: Integrate Phase-1 parallel subroutines into Barracuda and release (See Section Results from Phase-1 and its subsections) Task 6: Roadmap of Phase-2 (See Section Plan for Phase-2) With the completion of Phase 1 we have the base understanding to completely parallelize Barracuda. An overview of the work to move Barracuda to a parallelized code is given in Plan for Phase-2.

02 PETROLEUM↗

ChatHPC: Building the Foundations for a Productive and Trustworthy AI-Assisted HPC Ecosystem

ChatHPC democratizes large language models for the high-performance computing (HPC) community by providing the infrastructure, ecosystem, and knowledge needed to apply modern generative AI technologies to rapidly create specific capabilities for critical HPC components while using relatively modest computational resources. Our divide-and-conquer approach focuses on creating a collection of reliable, highly specialized, and optimized AI assistants for HPC based on the cost-effective and fast Code Llama fine-tuning processes and expert supervision. We target major components of the HPC software stack, including programming models, runtimes, I/O, tooling, and math libraries. Thanks to AI, ChatHPC provides a more productive HPC ecosystem by boosting important tasks related to portability, parallelization, optimization, scalability, and instrumentation, among others. With relatively small datasets (on the order of KB), the AI assistants, which are created in a few minutes by using one node with two NVIDIA H100 GPUs and the ChatHPC library, can create new capabilities with Meta’s 7-billion parameter Code Llama base model to produce high-quality software with a level of trustworthiness of up to 90% higher than the 1.8-trillion parameter OpenAI ChatGPT-4o model for critical programming tasks in the HPC software stack.

Young, Aaron [ORNL] (ORCID:0000000254484667)↗