N2A: a language and software tool for large-scale modeling.
Abstract not provided.
Rothganger, Fredrick↗
SEARCH · Search NASA
Results for “Large language models”
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.
Teaching AI when to care about gender
Natural Language Processing (NLP) is a branch of Artificial Intelligence (AI) concerned with solving language tasks by modeling large amounts of textual data. Some NLP techniques use word embeddings which are semantic models where machine learning (ML) is used to learn to cluster semantically related words by learning about word co-occurrences in the original training text. Unfortunately, these models tend to reflect or even exaggerate biases that are present in the training corpus. Here we describe the Word Embedding Navigator (WEN), which is a tool for exploring word embedding models. We examine a specific potential use case for this tool: interactive discovery and neutralization of gender bias in word embedding models, and compare this human-in-the-loop approach to reducing bias in word embeddings with a debiasing post-processing technique.
97 MATHEMATICS AND COMPUTING↗
VWC-BERT: Scaling Vulnerability–Weakness–Exploit Mapping on Modern AI Accelerators
Defending cybersystems needs accurate mapping of software and hardware vulnerabilities to generalized descriptions of weaknesses, and weaknesses to exploits. These mappings enable cyber defenders to build plans for effective defense and assessment of potential risks to a cybersystem. With close to 170k vulnerabilities, manual mapping is not a feasible option. However, automated mapping is challenging due to limited training data, computational intractability, and limitations in computational natural language processing. Tools based on breakthroughs in Transformer-based language models have been demonstrated to classify vulnerabilities with high accuracy. We make three key contributions in this paper: (1) We present a new framework, \VWCBERT, that augments the Transformer-based hierarchical multi-class classification framework of Das et al. (\textsc{V2W-BERT}) with the ability to map weaknesses to exploits. (2) We implement \VWCBERT~ on modern AI accelerator platforms using two data parallel techniques for the pre-training phase and demonstrate nearly linear speedups across NVIDIA and Graphcore accelerator platforms. We observe nearly linear speedups for up to 16 V100 and 8 A100 GPUs, and about 3.4$\times$ speedup for A100 relative to V100 GPUs. We also observe excellent speedups on Graphcore, with $5.7\times$ speedup on 128 IPUs relative to 16 IPUs. Enabled by scaling, we also demonstrate higher accuracy using a larger language model, RoBERTa-Large. We show up to 87\% accuracy for strict and up to 98\% accuracy for relaxed classification. (3) We develop a novel parallel link manager for the link prediction phase and demonstrate up to 21$\times$ speedup with 16 V100 GPUs relative to one V100 GPU, and thus reducing the runtime from 2.5 hours to 10 minutes. We believe that generalizability and scalability of \VWCBERT~ will benefit both the theoretical development and practical deployment of novel cyberdefense solutions and vulnerability classification.
Das, Siddhartha Shankar↗
pnnl/SciTune
Scientific multimodal instruction tuning with large language and vision models.
Horawalavithana, Sameera [Pacific Northwest Nation↗
Foundation models for atomistic simulation of chemistry and materials
Conventional computational methods for modeling chemical and materials systems are limited by system size and timescale, forcing a trade-off between quantum-mechanical accuracy and the sampling needed for realistic observables. Large language and vision foundation models — pre-trained on massive datasets using transformer architectures — have revolutionized many fields. It is thus interesting to ask whether a foundation model — subject to suitable data, parameter scaling and training — could enable learned simulations of chemistry and materials. Here, in this study, we review the field of machine-learned interatomic potentials (MLIPs) and posit that scaling up large and diverse chemical and materials datasets and highly expressive architectures using advanced training strategies should result in models that are: more efficient, transferable, robust to out-of-distribution scenarios, and easier to fine-tune to a variety of downstream physical observables than models trained from scratch on small datasets corresponding to specific, targeted atomistic simulation tasks. We provide specific criteria for creating such large-scale MLIP foundation models, coordinated strategies for their development, evaluation and deployment, and highlight potential emergent capabilities that could transform predictive simulations in chemistry and materials science and accelerate discovery across multiple technological domains.
Yuan, Eric C.-Y. [University of California, Berkel↗
TransPolymer: a Transformer-based language model for polymer property predictions
Accurate and efficient prediction of polymer properties is of great significance in polymer design. Conventionally, expensive and time-consuming experiments or simulations are required to evaluate polymer functions. Recently, Transformer models, equipped with self-attention mechanisms, have exhibited superior performance in natural language processing. However, such methods have not been investigated in polymer sciences. Herein, we report TransPolymer, a Transformer-based language model for polymer property prediction. Our proposed polymer tokenizer with chemical awareness enables learning representations from polymer sequences. Rigorous experiments on ten polymer property prediction benchmarks demonstrate the superior performance of TransPolymer. Moreover, we show that TransPolymer benefits from pretraining on large unlabeled dataset via Masked Language Modeling. Experimental results further manifest the important role of self-attention in modeling polymer sequences. We highlight this model as a promising computational tool for promoting rational polymer design and understanding structure-property relationships from a data science view.
36 MATERIALS SCIENCE↗
Accelerating Transformer-based Deep Learning Models on FPGAs using Column Balanced Block Pruning
Although Transformer-based language representations achieve state-of-the-art accuracy on various natural language processing (NLP) tasks, the large model size has been challenging the resource constrained computing platforms. Weight pruning, as a popular and effective technique in reducing the number of weight parameters and accelerating the Transformer, has been investigated on GPUs. However, the Transformer acceleration using weight pruning on field-programmable gate array (FPGAs) remains unexplored. This paper investigates the column balanced block-wise pruning on Transformer and designs an FPGA acceleration engine to customize the balanced blockwise matrix multiplication. We implement the Transformer model with proper hardware scheduling, and the experiments show that the Transformer inference on FPGA achieves 10.35 ms latency with the batch size of 32, which is 10.96 × speed up comparing to CPU platform and 2.08 × speed up comparing to GPU platform.
Peng, Hongwu↗
Adaptive language model training for molecular design
Abstract The vast size of chemical space necessitates computational approaches to automate and accelerate the design of molecular sequences to guide experimental efforts for drug discovery. Genetic algorithms provide a useful framework to incrementally generate molecules by applying mutations to known chemical structures. Recently, masked language models have been applied to automate the mutation process by leveraging large compound libraries to learn commonly occurring chemical sequences (i.e., using tokenization) and predict rearrangements (i.e., using mask prediction). Here, we consider how language models can be adapted to improve molecule generation for different optimization tasks. We use two different generation strategies for comparison, fixed and adaptive. The fixed strategy uses a pre-trained model to generate mutations; the adaptive strategy trains the language model on each new generation of molecules selected for target properties during optimization. Our results show that the adaptive strategy allows the language model to more closely fit the distribution of molecules in the population. Therefore, for enhanced fitness optimization, we suggest the use of the fixed strategy during an initial phase followed by the use of the adaptive strategy. We demonstrate the impact of adaptive training by searching for molecules that optimize both heuristic metrics, drug-likeness and synthesizability, as well as predicted protein binding affinity from a surrogate model. Our results show that the adaptive strategy provides a significant improvement in fitness optimization compared to the fixed pre-trained model, empowering the application of language models to molecular design tasks.
37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
Chemical classification program synthesis using generative artificial intelligence
Accurately classifying chemical structures is essential for cheminformatics and bioinformatics, including tasks such as identifying bioactive compounds of interest, screening molecules for toxicity to humans, finding non-organic compounds with desirable material properties, or organizing large chemical libraries for drug discovery or environmental monitoring. However, manual classification is labor-intensive and difficult to scale to large chemical databases. Existing automated approaches either rely on manually constructed classification rules, or are deep learning methods that lack explainability. This work presents an approach that uses generative artificial intelligence to automatically write chemical classifier programs for classes in the Chemical Entities of Biological Interest (ChEBI) database. These programs can be used for efficient deterministic run-time classification of SMILES structures, with natural language explanations. The programs themselves constitute an explainable computable ontological model of chemical class nomenclature, which we call the ChEBI Chemical Class Program Ontology (C3PO). We validated our approach against the ChEBI database, and compared our results against deep learning models and a naive SMARTS pattern based classifier. C3PO outperforms the naive classifier, but does not reach the performance of state of the art deep learning methods. However, C3PO has a number of strengths that complement deep learning methods, including explainability and reduced data dependence. C3PO can be used alongside deep learning classifiers to provide an explanation of the classification, where both methods agree. The programs can be used as part of the ontology development process, and iteratively refined by expert human curators.
Artificial Intelligence↗
Genomic Language model for Annotation of Repetitive Elements (GLARE) v1.0
GLARE (Genomic Language model for Annotation of Repetitive Elements) is a tool that classifies transposable elements (TEs)—the mobile, repetitive DNA sequences that make up large fractions of eukaryotic genomes. GLARE fine-tunes the NTv3-650M genomic language model on a harmonized collection of curated TE sequences from the PanTEon and Repbase reference databases, assigning each input sequence to one of 11 orders and 32 superfamilies in a Wicker-compatible taxonomy. Features. From nucleotide FASTA input, GLARE outputs per-sequence predictions, class summaries, composition figures, and an annotated FASTA. It provides calibrated confidence scores with optional abstention and runs on CPU or GPU. Uses. GLARE serves as a classification component in genome-annotation pipelines, downstream of TE discovery, supporting genome annotation and comparative and evolutionary genomics. Advantages. GLARE is the first repeat-element classifier to leverage a pretrained genomic language model. Combined with multi-database training, this approach outperformed all nine classifiers in the PanTEon benchmark, generalized better to unseen taxonomic clades, and remained robust to sequence orientation—a common failure mode of existing tools.
Bruna, Tomas [Lawrence Berkeley National Laborator↗
Foundation Models of Scientific Knowledge for Chemistry: Opportunities, Challenges and Lessons Learned
Foundation models pre-trained on large corpora demonstrate significant gains across many natural language processing tasks and domains e.g., law, healthcare, education, etc. However, only limited efforts have investigated the opportunities and limitations of applying these powerful models to science and security applications. In this work we develop foundation models of scientific knowledge for chemistry to augment scientists with the advanced ability to perceive and reason at scale previously unimagined. Specifically, we build large-scale (1.47B parameter) general-purpose models for chemistry that can be effectively used to perform a wide range of in-domain and out-of-domain tasks. Evaluating these models in a zero-shot setting, we analyze the effect of model and data scaling, knowledge depth, and temporality on model performance in context of model training efficiency. Our novel findings demonstrate that (1) model size significantly contributes to the task performance when evaluated in a zero-shot setting; (2) data quality (aka diversity) affects model performance more than data quantity; (3) similarly, unlike previous work (Luu et al., 2021) temporal order of the documents in the corpus boosts model performance only for specific tasks, e.g., SciQ; and (4) models pre-trained from scratch perform better on in-domain tasks than those tuned from general-purpose models like Open AI’s GPT-2.
Foundation Models, Chemistry↗
Agentic artificial intelligence for multistage physics experiments at a large-scale user facility particle accelerator
We present a language-model-driven agentic artificial intelligence (AI) system to autonomously execute multistage physics experiments on a production synchrotron light source. Implemented at the Advanced Light Source particle accelerator, the system translates natural language user prompts into structured execution plans that combine archive data retrieval, control-system channel resolution, automated script generation, controlled machine interaction, and analysis. In a representative machine physics task, we show that preparation time was reduced by 2 orders of magnitude relative to manual scripting even for a system expert, while operator-standard safety constraints were strictly upheld. Core architectural features, plan-first orchestration, bounded tool access, and dynamic capability selection, enable transparent, auditable execution with fully reproducible artifacts. These results establish a blueprint for the safe integration of agentic AI into accelerator experiments and demanding machine physics studies, as well as routine operations, with direct portability across accelerators worldwide and, more broadly, to other large-scale scientific infrastructures.
Accelerator/storage ring control systems↗
System-Level Integration of Modular Language Models for Real-Time Risk Assessment in Third-Party Risk Management Systems
Large enterprises typically rely on dedicated teams to govern and implement security measures throughout their supply chains, ensuring compliance with enterprise security procedures. There is a significant reliance on Third-Party Risk Management (TPRM) platforms, which often require complete, highly structured information from potential vendors. The review and compliance assurance processes are time- and labor intensive, often requiring several rounds of review between the supply chain security risk management teams, business users, and potential vendors, leading to delays in the supply chain processing and consumer experience. Significant challenges in the risk management paradigm include handling unstructured data in various formats and providing real-time feedback to users to reduce the required review time. This paper presents a novel solution to these challenges. A modular multi-step system architecture is proposed using advances in language processing, specifically for unstructured responses and provides real-time feedback (i.e., 3 seconds) so that users can improve their responses before the TPSRM team review. This novel system architecture will increase information accuracy and significantly reduce time and labor during the review process.
99 - GENERAL AND MISCELLANEOUS↗
A case study in contrastive learning information combination: Application to technical forensics of additive manufacturing filament source identification
Combination of information from disparate data sources into a single decision is a core challenge in many fields, including the field of technical forensics. Technical forensics (TF) utilizes technical characterization of questioned samples to determine properties of that sample; these properties are then used to infer information of forensic interest, such as provenance, age, or attribution. TF is utilized in traditional forensic applications, such as the attribution of material fragments from an explosive, and in nuclear forensic applications, such as the attribution of actinides which have been interdicted out of regulatory control. The challenge of combining information from disparate sources, described alternately by many terms including “Data Fusion” and “Data Integration”, is exacerbated in the technical forensics domain due to at least two factors: the challenge of interpreting each information source singularly, and the relatively small data set sizes available. Extensive literature exists attempting to combine technical forensics information sources, both in manual and automated processes. These attempts are often bespoke to the specific information sources (such as the bi-, tri-, or quad-isotope chart (Moody, Grant, and Hutcheon 2005)), with some emerging examples of simple early- and late- fusion (, respectively). Simultaneous to the information combination efforts described in the previous paragraph, the field of natural language processing attempted (and largely succeeded) in combining information from multiple non-technical information sources. The ecosystem of “multi-modal” language models, which can take text and images as input, and generate text and images as output, became large and diverse by 2025 (Khan et al. 2025). In a generalized sense, many of these methods are trained by learning neural networks which can convert raw text or images into a vector of numbers describing the text or image, hereafter called “embeddings” and the neural networks performing the conversion are called “embedders”. By using a separate embedder for text and images, finding coincident text and images (such as images with their captions), and optimizing the parameters of the embedders such that the embeddings for the text and the image are similar, the field has found a bridge between text and images (Girdhar et al. 2023). It is the contention of the authors of this report that this insight is not limited to text and images but instead can be extended to any modality which can be found coincidently. The subject of the rest of this report is the application of this method to example multi-modal technical forensic data. Some details about the data used in this report are not appropriate for this report, and are included in a companion report (PNNL-38669).
36 MATERIALS SCIENCE↗
Pretraining Billion-Scale Geospatial Foundational Models on Frontier
As AI workloads increase in scope, generalization capability becomes challenging for small task-specific models and their demand for large amounts of labeled training samples increases. On the contrary, Foundation Models (FMs) are trained with internet-scale unlabeled data via self-supervised learning and have been shown to adapt to various tasks with minimal fine-tuning. Although large FMs have demonstrated significant impact in natural language processing and computer vision, efforts toward FMs for geospatial applications have been restricted to smaller size models, as pretraining larger models requires very large computing resources equipped with state-of-the-art hardware accelerators. Current satellite constellations collect 100+TBs of data a day, resulting in images that are billions of pixels and multimodal in nature. Such geospatial data poses unique challenges opening up new opportunities to develop FMs. We investigate billion scale FMs and HPC training profiles for geospatial applications by pretraining on publicly available data. We studied from end-to-end the performance and impact in the solution by scaling the model size. Our larger 3B parameter size model achieves up to 30% improvement in top1 scene classification accuracy when comparing a 100M parameter model. Moreover, we detail performance experiments on the Frontier supercomputer, America's first exascale system, where we study different model and data parallel approaches using PyTorch's Fully Sharded Data Parallel library. Specifically, we study variants of the Vision Transformer architecture (ViT), conducting performance analysis for ViT models with size up to 15B parameters. By discussing throughput and performance bottlenecks under different parallelism configurations, we offer insights on how to leverage such leadership-class HPC resources when developing large models for geospatial imagery applications.
Tsaris, Aristeidis (aris)↗
Towards a Techno-Economic Analysis of PCM Integrated Hybrid HVAC Systems
Thermal end uses dominate building energy consumption and are a major driver of peak demand. As heating is electrified, peak electrical power required will surge, prompting a need for innovative HVAC system designs and controls.These designs must incorporate novel technologies at the component level and new integration techniques at the system level. One such possibility involves the addition of thermal energy storage (TES) in heating and cooling equipment using a phase change material (PCM) heat exchanger. Here, thermal energy storage via phase change can be used to shift the HVAC system loads to times of lower electricity cost, reduced carbon intensity, and greater energy efficiency. Most of the current utilization of PCM in buildings involves passive components. By actively controlling when heat is stored and released from PCM, we can optimize the building HVAC system to cost-effectively meet consumer needs with the flexibility to draw on renewable energy resources when they are abundant and available. While this combination of technologies is promising in theory, simulation-based evaluation of a prototype can be difficult due to the modeling requirements at the component level and the large number of possible configurations and operating modes at the system level. To conduct this evaluation, we use the Modelica language for modeling and simulation because it enables users to represent the important physics of the problem, interchange and rearrange components in an efficient manner, and implement a range of control configurations. In this work, we considered three case studies: a portable building, a large commercial retail store, and a multifamily residential apartment unit. Each of these employs a different system design, ranging from a single package vertical unit incorporating PCM to a central plant with independent heat pump, evaporative cooling, and thermal energy storage components. This paper describes the technologies in question, presents modeling at the component and system levels, and demonstrates building energy and demand charge cost savings with local time-of-use tariffs in a hot-dry climate.
Helmns, Dre↗
Autogenerating a Domain-Specific Question-Answering Data Set from a Thermoelectric Materials Database to Enable High-Performing BERT Models
We present a method for autogenerating a large domain-specific question-answering (QA) dataset from a thermoelectric materials database. We show that a small language model, BERT, once fine-tuned on this automatically generated dataset of 99,757 QA pairs about thermoelectric materials, affords better performance in the field of thermoelectric materials compared to a BERT model fine-tuned on the generic English-language QA data set, SQuAD-v2. We further show that mixing the two data sets (ours and SQuAD-v2), which have significantly different syntactic and semantic scopes, allows the BERT model to achieve even better performance. The best-performing BERT model fine-tuned on the mixed data set outperforms the models fine-tuned on the other two data sets by scoring an exact match of 67.93% and an F1 score of 72.29% when evaluated on our test data set. This has important implications as it demonstrates the ability to realize high-performing small language models, with modest computational resources, empowered by domain-specific materials data sets which can be generated according to our method.
biological databases↗