SEARCH · Engineering Papers
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.
This presentation is an inventory of the latest large language model (LLM) offerings. It lists some of the most prominent open- and closed-source LLMs. Given the rapid pace of progress in this space, it is not guaranteed to be a comprehensive list. This resource is intended to help inform decision-making about LLM capabilities at NREL.
ProtAgents is a de novo protein design platform based on multimodal LLMs, where distinct AI agents with expertise in knowledge retrieval, protein structure analysis, physics-based simulations, and results analysis tackle tasks in a dynamic setting.
The goal of this mini report is to give opinions on the current state of the art in large language model (LLM) assisted code translation, which aims to replicate the behavior of a program written in one programming language to another. The first section provides our overall sense of literature and suggested directions of research, and the second section summarizes a few of the most relevant papers in greater depth.
A major weakness of a Large Language Model (LLM) is its tendency to accept information at face value, often leading to injection of erroneous information and inducing a greater probability of hallucinating non-existent information. While Retrieval Augmented Generation (RAG) uses external knowledge sources to bolster LLMs through grounded truth, this work seeks to explore methods to engender a LLM with an intrinsic capability to evaluate an input’s believability without relying on external knowledge sources. We investigate unifying a LLM with a Knowledge Graph (KG) and using the KG to reinforce the LLM’s internal word embedding while also maintaining belief metrics along the edge’s in the KG.
Discovering novel catalysts requires complex reasoning involving multiple chemical properties and resultant trade-offs, leading to a combinatorial growth in the search space. While large language models (LLM) have demonstrated novel capabilities for chemistry through complex instruction following capabilities and high quality reasoning, a goal-driven combinatorial search using LLMs has not been explored in detail. In this work, we present a Monte Carlo Tree Search-based approach that improves beyond state-of-the-art chain-of-thought prompting variants to augment scientific reasoning. We introduce two new reasoning datasets: 1) a curation of computational chemistry simulations, and 2) diverse questions written by catalysis researchers for reasoning about novel chemical conversion processes. We improve over the best baseline by 25.8\% and find that our approach can augment scientist's reasoning and discovery process with novel insights.\footnote{All resources will be publicly available upon publication.
Publicly funded organizations, notably institutions like the Los Alamos National Laboratory (LANL), are deeply vested in acquiring robust productivity metrics to gauge the entirety of their research output. Motivated by the imperative to enhance institutional productivity assessment, this study investigates the utilization of Large Language Models (LLM) such as BERT-based models, as well as local LLaMa-30b-instruct and Mixtral-8x7b-instruct architectures for classifying type of URL referenced resources in academic papers such as software, dataset, as well as authorship intent. Challenges in discerning resource types from context are highlighted, along with the potential of BERT and LLMs to address these challenges. Through comprehensive analysis, this research unveils a notable surge in documents featuring URL citations, indicative of the escalating importance of digital resources in scholarly publications. Moreover, citations to datasets and software demonstrate consistent growth over time, underscoring their increasing significance. Our findings also reveal that LANL authors contribute substantially to accessible science, comprising about 10% of dataset and software mentions in LANL
We apply AI-assisted large language model (LLM) capabilities of GPT-3 targeting high-performance computing (HPC) kernels for (i) code generation, and (ii) auto-parallelization of serial code in C ++, Fortran, Python and Julia. Our scope includes the following fundamental numerical kernels: AXPY, GEMV, GEMM, SpMV, Jacobi Stencil, and CG, and language/programming models: (1) C++ (e.g., OpenMP [including offload], OpenACC, Kokkos, SyCL, CUDA, and HIP), (2) Fortran (e.g., OpenMP [including offload] and OpenACC), (3) Python (e.g., numpy, Numba, cuPy, and pyCUDA), and (4) Julia (e.g., Threads, CUDA.jl, AMDGPU.jl, and KernelAbstractions.jl). Kernel implementations are generated using GitHub Copilot capabilities powered by the GPT-based OpenAI Codex available in Visual Studio Code given simple + + prompt variants. To quantify and compare the generated results, we propose a proficiency metric around the initial 10 suggestions given for each prompt. For auto-parallelization, we use ChatGPT interactively giving simple prompts as in a dialogue with another human including simple “prompt engineering” follow ups. Results suggest that correct outputs for C++ correlate with the adoption and maturity of programming models. For example, OpenMP and CUDA score really high, whereas HIP is still lacking. We found that prompts from either a targeted language such as Fortran or the more general-purpose Python can benefit from adding language keywords, while Julia prompts perform acceptably well for its Threads and CUDA.jl programming models. Finally, we expect to provide an initial quantifiable point of reference for code generation in each programming model using a state-of-the-art LLM. Overall, understanding the convergence of LLMs, AI, and HPC is crucial due to its rapidly evolving nature and how it is redefining human-computer interactions.
The COVID-19 pandemic demonstrated how a novel, elusive, and diffuse biological threat can engender uncertainty and misinformation, and it underscored the need for flexible analytical modalities agnostic to the identity of biological material. Yet even before the pandemic, recognition of the limitations of the current, list-based approach, which focuses on known pathogens and biotoxins, and of the importance of agent-agnostic biodetection was growing within the biosecurity community. In a 2018 report on “Biodefense in the Age of Synthetic Biology,” for example, the National Academy of Sciences stated that “an overreliance on the Select Agent List is a systematic weakness affecting many aspects of the United States’ current biodefense mitigation capability”. More recently, a group of biodefense researchers proposed the identification and adoption of “bioagent-agnostic signatures (BASs)” as a way of detecting and characterizing not only existing agents but also novel ones, an approach they believe will “enable a more flexible and resilient biodefense posture”. Indeed, the future of biodetection requires us to begin developing novel analytics that can identify anomalies and/or characteristics that indicate a potential threat, whether known or unknown, without looking for a specific signature that has been identified previously. To assess potential threats more rapidly, it is critical to develop agnostic artificial intelligence (AI)/machine learning (ML) systems that can be employed for real-time assessment of the nature and source of a perturbation. Such systems should be multi scale and multi-dimensional, integrating sensor data from a range of biological, chemical, and physical application spaces. Emerging deep learning (DL) models demonstrate exceptional promise for identification of discriminatory features within multi-dimensional datasets. DL models have the capacity to recognize and encode highly complex patterns in a wide range of input data modalities, including images, text, and biological/chemical/physical spectra. As such, they can execute a wide range of assessments and determinations that have traditionally required a human operator. The promise of advances in DL is apparent in the realm of human health and medicine. DL models have been validated for evaluating a variety of clinical threats to human health in a range of contexts, including infection and cancer, and they demonstrated improved performance in predicting stroke relative to human neurologists in some categories of data. Continuously evolving advances in AI/ML are expected to support more efficient evaluation of raw sequence, spectroscopy, and spectrometry data. For instance, recent advances and deployment of large language models (LLM) such as Generative Pre training Transformer (GPT) and Bidirectional Encoder Representations from Transformers (BERT) have already motivated application of these models for biological function prediction. As frameworks such as LLMs become larger and more complex in their representations, their capacity to serve as pre-trained models that can be fine-tuned for biological/biodetection purposes will similarly be amplified. While existing and emerging AI/ML have found broad applicability and use cases in the clinical sciences, development for environmental evaluation and biodetection has been limited. Functionalizing such capabilities for this purpose requires an understanding of the existing technical landscape and how the respective tools and algorithms are currently being employed. This landscape awareness then allows an assessment of the current practical capabilities of existing models and the anticipated requirements and development efforts that will be needed to adapt available algorithms for biodetection applications relevant to DHS. Leveraging expertise in biodetection, ML, and operational biodetection, the effort described in this report is comprised of a systematic landscape assessment (Subtask 2.1), comparative evaluation (Subtask 2.2), and formulation of a value proposition (Subtask 2.3) for the prospect of ML-enabled, agnostic biodetection from raw, or minimally-processed, datasets.
Here, we present an autonomous materials discovery framework that couples a large language model (LLM) with molecular dynamics (MD) simulations to optimize Fe–Cr–Mn alloy compositions for tensile strength. Starting from six distinct compositions, the LLM operated as an intelligent agent, iteratively proposing changes based on prior simulation results and constraints. Over 50 iterations per case, the LLM adaptively explored the composition space, identifying high-strength regions, not easily accessible by conventional methods. The highest strength, 18.7 GPa, was achieved with Fe 71 Cr 25 Mn 4 composition, identified from a Fe 75 Cr 20 Mn 5 starting point. The LLM autonomously adjusted its strategy in real time, demonstrating closed-loop decision-making using commodity hardware. This approach showcases the potential of LLMs as scientific co-pilots, capable of accelerating materials discovery and generalizable to other domains like biology and drug design.
With advances in imaging technologies, modern photovoltaic (PV) systems generate large volumes of heterogeneous image data, including visible, electroluminescence (EL), and infrared (IR) images. Existing PV image analysis models, particularly deep learning approaches, are typically task-specific and lack cross-modality generalization. To address this limitation, this paper proposes an open-source large language model (LLM)–based unified framework for heterogeneous PV image diagnostics. Through task-aware diagnostic prompting, the framework enables analysis of visible, EL, and IR images within a single pipeline, supporting both zero-shot and few-shot inference and binary and multiclass classification. It is compatible with state-of-the-art multimodal LLMs, including ChatGPT, Gemini, Claude, Qwen, and CLIP. The framework is evaluated on PV module condition classification (clean, soiling, snow, hail, and bird droppings) using visible images, cell crack detection using EL images, and hotspot detection using IR images. GPT-5.1 in few-shot mode achieves the best performance, with classification accuracy exceeding 97.3%. Open-source models such as Qwen and CLIP also deliver competitive results on visible images (around 90% accuracy), though their performance is more limited on EL and IR modalities. On the full ELPV dataset, the framework achieves 83.5% zero-shot accuracy, within 2.8% of the supervised CNN baseline, confirming scalability to larger benchmarks. Practical aspects such as reproducibility, response latency, and confidence estimation are systematically analyzed. The framework operates across PV image modalities without modality- or task-specific training, making it well suited as a rapid pre-screening tool to support downstream detailed diagnostics. A benchmark dataset of diverse labeled PV images is also released.
Deep learning models have been shown to be vulnerable to adversarial attacks, in which perturbations to their inputs cause the model to produce incorrect predictions. As opposed to adversarial attacks in computer vision, where small changes introduced to pixel values can drastically alter a model's output while remaining imperceptible to humans, text-based attacks are difficult to conceal due to the discrete nature of tokens. Consequently, unconstrained gradient-based attacks often produce adversarial examples that lack semantic meaning, rendering them detectable through visual inspection or perplexity filters. In contrast to methods that rely on gradient-based optimization in the embedding space, we propose an approach that leverages a Large Language Model's ability to generate grammatically correct and semantically meaningful text to craft adversarial patches that seamlessly blend in with the original input text. These patches can be used to alter the behavior of a target model, such as a text classifier. Since our approach does not rely on gradient backpropagation, it only requires access to the target model's confidence scores, making it a grey-box attack. We demonstrate the feasibility of our approach using open-source LLMs, including Intel's Neural Chat, Llama2, and Mistral-Instruct, to generate adversarial patches capable of altering the predictions of a distilBERT model fine-tuned on the IMDB reviews dataset for sentiment classification.
The Message Passing Interface (MPI) standard plays a crucial role in enabling scientific applications for parallel computing and is an essential component in high-performance computing (HPC). However, implementing MPI code manually—especially applying a proper domain decomposition and communication pattern—is a challenging and error-prone task. We present ChatMPI, an AI assistant for MPI parallelization of sequential C codes. In our analysis, we focus on testing six essential HPC workloads, which are based on Basic Linear Algebra Subprograms levels 1, 2, and 3 as well as sparse, stencil, and iterative operations. We analyze the process of creating ChatMPI by using the ChatHPC library. This lightweight large language model (LLM)–based infrastructure enables HPC experts to efficiently create and supervise trustworthy AI capabilities for critical HPC software tasks. We study the data required for training (fine-tuning) ChatMPI to generate parallel codes that not only use MPI syntax correctly but also apply HPC techniques to reduce memory communication and maximize performance by using proper work decomposition. With a relatively small training dataset composed of a few dozen prompts and fewer than 15 minutes of fine-tuning on one node equipped with two NVIDIA H100 GPUs, ChatMPI elevates trustworthiness for MPI code generation of current LLMs (e.g., Code Llama, ChatGPT-4o and ChatGPT 5). Additionally, we evaluate the performance of the MPI codes generated by ChatMPI in comparison with the ones generated by ChatGPT-4o and ChatGPT-5. The codes generated by ChatMPI provide up to a 4 × boost in performance by using better problem decomposition, communication patterns, and HPC techniques (e.g., communication avoiding).
Deep neural networks (DNNs) have achieved tremendous success in the past few years. However, their training and inference demand exceptional computational and memory resources. Quantization has been shown as an effective approach to mitigate the cost, with the mainstream data types reduced from FP32 to FP16/BF16 and recently FP8 in the latest NVIDIA H100 GPUs. With increasingly aggressive quantization, however, the conventional floating-point formats suffer from limited precision in representing numbers around zero. Recently, NVIDIA demonstrated the potential of using a Logarithmic Number System (LNS) for the next generation of tensor cores. While LNS mitigates the hurdles in representing small numbers, in this work we observed a mismatch between LNS and the emerging Large Language Models (LLM), where LLM exhibits significant outliers when directly adopting the LNS format. In this paper, we present a data-format/architecture codesign to bright this gap. On the format side, we propose a dynamic LNS format to flexibly represent outliers at a higher precision, by exploiting asymmetry in the LNS representation and identifying outliers through a per-vector basis. On the architecture side, for demonstration, we realize the dynamic LNS format in a systolic array, which can handle the irregularity of the outliers at runtime. We implement our approach on an Alveo U280 FPGA as a prototype. Experimental results show that our design can effectively handle the outliers and resolve the mismatch between LNS and LLM, contributing to an accuracy improvement of 15.4% and 16% over the floating-point and the original LNS baselines, using four state-of-the-art LLM models. Our observation and design lay a solid foundation for the large-scale adoption of the LNS format in the next-generation deep learning hardware.
Production LLM serving must simultaneously deliver high throughput, low latency, and sufficient context capacity under non-stationary traffic and mixed request requirements. Data parallelism (DP) maximizes throughput by running independent replicas, while tensor parallelism (TP) reduces per-request latency and pools memory for long-context inference. However, existing serving stacks typically commit to a static parallelism configuration at deployment; adapting to bursts, priorities, or long-context requests is often disruptive and slow. We present Flying Serving, a vLLM-based system that enables online DP-TP switching without restarting engine workers. Flying Serving makes reconfiguration practical by virtualizing the state that would otherwise force data movement: (i) a zero-copy Model Weights Manager that exposes TP shard views on demand, (ii) a KV Cache Adaptor that preserves request KV state across DP/TP layouts, (iii) an eagerly initialized Communicator Pool to amortize collective setup, and (iv) a deadlock-free scheduler that coordinates safe transitions under execution skew. Across three popular LLMs and realistic serving scenarios, Flying Serving improves performance by up to 4.79 × under high load and 3.47 × under low load while supporting latency- and memory-driven requests.
FAIR to WISE (F2W) is an iterative, large-language model (LLM) driven pipeline that turns unstructured research PDFs into structured, queryable knowledge graphs (KGs). Core features include schema-driven extraction to a LinkML model; full provenance capture; ontology-grounded enrichment (e.g., chemical validation and ChEBI lookup); graph construction to JSON-LD with stable IDs; and KG-RAG question answering with evidence-aware retrieval. The system is engineered for reproducibility and accessibility (open-source Ollama models, temperature=0, NVTX/Nsight profiling) with robust QA (relation verification, deduplication, and deterministic outputs). Primary uses are literature-to-KG automation, knowledge-grounded Q&A, and experimental steering support. We demonstrate the approach in organic photovoltaics, where the pipeline ingests papers, builds a domain KG, and evaluates answers against expert competency questions to guide experimental planning and interpretation. Compared with off-the-shelf LLMs and ad-hoc NLP tools, F2W addresses ontology gaps and reduces hallucination risk by grounding responses in extracted evidence and enforcing schema constraints; it also offers deterministic, provenance-linked outputs and open, cost-aware deployment. Evidence-aware ranking further improves answer quality over pure vector search.
Emerging operational concepts for aviation hinge on novel paradigms for human machine interaction. Critical to their safe operation is early consideration of human error into the design process. Existing methods for consideration of human error require significant expert input, which is challenging both in early design and in novel systems for which there is little existing safety expertise. In this research, we propose a methodology for identifying human error, error producing factors, and mechanisms in early design from historical incident reports. Additionally, we hypothesize that cross-domain sharing of lessons learned can aid with early design human considerations in circumstances where data is not relevant or incomplete. This is addressed by identifying causes of human error in aviation and railway domains through applying state-of-the art natural language processing techniques to historical incident reports. Using this method, it is possible to extract extensive reports on human error from past incidents. Using the proposed approach, we identify nine human errors from railway reports and fourteen from aviation reports, with three errors common to both domains. There is at least one error producing conditions for each human error while a majority of the errors have more than one error mechanism. We also found that a majority of the human errors, error producing factors, and error mechanisms (even if they are not common between the domains) can be used to inform safe operations across domains as long as the errors are not domain specific and are interpreted and contextualized using engineering judgement.