Search NASA⌕ Search

SEARCH · Search NASA

Results for “Generative Pretrained Transformer”

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.

14 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

A survey of techniques for optimizing transformer inference

Recent years have seen a phenomenal rise in the performance and applications of transformer neural networks. The family of transformer networks, including Bidirectional Encoder Representations from Transformer (BERT), Generative Pretrained Transformer (GPT) and Vision Transformer (ViT), have shown their effectiveness across Natural Language Processing (NLP) and Computer Vision (CV) domains. Transformer-based networks such as ChatGPT have impacted the lives of common men. However, the quest for high predictive performance has led to an exponential increase in transformers' memory and compute footprint. Researchers have proposed techniques to optimize transformer inference at all levels of abstraction. Further, this paper presents a comprehensive survey of techniques for optimizing the inference phase of transformer networks. We survey techniques such as knowledge distillation, pruning, quantization, neural architecture search and lightweight network design at the algorithmic level. We further review hardware-level optimization techniques and the design of novel hardware accelerators for transformers. We summarize the quantitative results on the number of parameters/FLOPs and the accuracy of several models/techniques to showcase the tradeoff exercised by them. We also outline future directions in this rapidly evolving field of research. We believe that this survey will educate both novice and seasoned researchers and also spark a plethora of research efforts in this field.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Using Generative AI to implement the discrepancy checker for a Nearly Autonomous Management and Control System for Advanced Reactors

Developments related to generative artificial intelligence (AI) have brought a major breakthrough in AI. These developments are rapidly accelerating developments in different science and engineering applications. Nearly Autonomous Management and Control (NAMAC) system provides recommendations to the operator for maintaining the safety and performance of the reactor. The discrepancy checker (DC) is an important component of the NAMAC) system, whose goal is to determine if the plant is moving towards the expected system state after the control actions are injected. In this work, we explore generative AI methods, particularly, a generative pretrained transformer (GPT) for implementing the DC function in NAMAC. The GPT-based DC aims to alert the operator in situations outside NAMAC’s scope and act as a chatbot the operator can use to retrieve relevant information. This study involves two versions of GPT developed by OpenAI: GPT-3.5 and GPT-4. These GPTs are trained on huge amounts of undisclosed general domain datasets. We explored two methods to adapt GPTs for DC implementation in NAMAC: fine-tuning and retrieval augmented generation. A small knowledge base (information file) that encompasses rules for DC implementation and some general information related to NAMAC has been created to support DC implementation using GPT. In this work, the GPT-based DC implementations have been tested for their reasoning abilities, comprehension, information retrieval, and extraction abilities. It should be noted that this paper only presents a preliminary study to test the feasibility of DC implementation using generative AI technology. Given the potential risks and severe consequences associated with nuclear reactor applications, combined with the black-box nature of AI, extensive offline and online testing and reliability analyses of GPT-based DCs are needed for further developing such capabilities.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2025

Consistent performance of large language models in rare disease diagnosis across ten languages and 4917 cases

Background Large language models (LLMs) are increasingly used medicine for diverse applications including differential diagnostic support. The training data used to create LLMs such as the Generative Pretrained Transformer (GPT) predominantly consist of English-language texts, but LLMs could be used across the globe to support diagnostics if language barriers could be overcome. Initial pilot studies on the utility of LLMs for differential diagnosis in languages other than English have shown promise, but a large-scale assessment on the relative performance of these models in a variety of European and non-European languages on a comprehensive corpus of challenging rare-disease cases is lacking. Methods We created 4917 clinical vignettes using structured data captured with Human Phenotype Ontology (HPO) terms with the Global Alliance for Genomics and Health (GA4GH) Phenopacket Schema. These clinical vignettes span a total of 360 distinct genetic diseases with 2525 associated phenotypic features. We used translations of the Human Phenotype Ontology together with language-specific templates to generate prompts in English, Chinese, Czech, Dutch, French, German, Italian, Japanese, Spanish, and Turkish. We applied GPT-4o, version gpt-4o-2024-08-06, and the medically fine-tuned Meditron3-70B to the task of delivering a ranked differential diagnosis using a zero-shot prompt. An ontology-based approach with the Mondo disease ontology was used to map synonyms and to map disease subtypes to clinical diagnoses in order to automate evaluation of LLM responses. Findings For English, GPT-4o placed the correct diagnosis at the first rank 19.9% and within the top-3 ranks 27.0% of the time. In comparison, for the nine non-English languages tested here the correct diagnosis was placed at rank 1 between 16.9% and 20.6%, within top-3 between 25.4% and 28.6% of cases. The Meditron3 model placed the correct diagnosis within the first 3 ranks for 20.9% of cases in English and between 19.9% and 24.0% for the other nine languages. Interpretation The differential diagnostic performance of LLMs across a comprehensive corpus of rare-disease cases was largely consistent across the ten languages tested. This suggests that the utility of LLMs in clinical settings may extend to non-English clinical settings.

Artificial intelligence↗
DOE OSTI2024

eGridGPT: Trustworthy AI in the Control Room

This report outlines the synergy between human decision making and generative artificial intelligence (GenAI), where GenAI supports power system operators by analyzing procedures, suggesting actions, simulating scenarios with physics-based digital twins, and recommending optimal decisions. This report is the first research effort to apply large language models (LLMs), a type of GenAI, in the power grid control room. The authors describe the Electric Grid Generative Pretrained Transformer (eGridGPT), an LLM that virtually assists system operators. Developed with cybersecurity and regulatory requirements in mind, eGridGPT represents an opportunity to responsibly evolve control room technologies to meet the needs of a rapidly changing grid. As an innovative concept, eGridGPT seeks to spearhead productive discussions about the advanced technologies in the control room of the future amid the transition to clean energy.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2025

A Risk-Informed Approach to Trustworthiness Assessment in Digital Twins-Based Autonomous Control

In autonomous control systems, digital twins (DTs) are used to perform diagnostic and prognostic functions. The trustworthiness of these DTs is dependent on quality and coverage of the training data, model accuracy and integrity of sensor data. This work introduces a methodology to determine the trustworthiness of a DT system given faulty sensor data using a risk informed approach. Bayesian Belief Networks (BBNs) are used to propagate uncertainties and determine the probability of trustable recommendations. The decision to trust the control action provided by the DT is based on the DT output, expert opinion, and severity of problems. The performance of DTs is reliant on the data they are trained on. When they encounter out of distribution data, the trustworthiness of the recommendations decreases. To address this issue, we include an expert component that provides input on sensor degradation. For this, we utilize a generative artificial intelligence (AI) model, such as Generative Pretrained Transformer (GPT). The GPT functions as an expert with broad knowledge. The GPT is fine-tuned to understand and discriminate sensor degradation scenarios using manufactured data. This methodology is demonstrated through a case study on a Nearly Autonomous Management and Control System (NAMAC) during a steady state scenario. Various sensor degradation types with different severity levels are considered. Degraded sensor data is processed by the DT system and the fine-tuned GPT. Finally, using the BBN, we combine the GPT information and the DT output with its sources of uncertainty. This provides an output regarding the trustworthiness of the DT recommendation.

22 - GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2024

An evaluation of GPT models for phenotype concept recognition

Clinical deep phenotyping and phenotype annotation play a critical role in both the diagnosis of patients with rare disorders as well as in building computationally-tractable knowledge in the rare disorders field. These processes rely on using ontology concepts, often from the Human Phenotype Ontology, in conjunction with a phenotype concept recognition task (supported usually by machine learning methods) to curate patient profiles or existing scientific literature. With the significant shift in the use of large language models (LLMs) for most NLP tasks, we examine the performance of the latest Generative Pre-trained Transformer (GPT) models underpinning ChatGPT as a foundation for the tasks of clinical phenotyping and phenotype annotation. The experimental setup of the study included seven prompts of various levels of specificity, two GPT models (gpt-3.5-turbo and gpt-4.0) and two established gold standard corpora for phenotype recognition, one consisting of publication abstracts and the other clinical observations. The best run, using in-context learning, achieved 0.58 document-level F1 score on publication abstracts and 0.75 document-level F1 score on clinical observations, as well as a mention-level F1 score of 0.7, which surpasses the current best in class tool. Without in-context learning, however, performance is significantly below the existing approaches. Our experiments show that gpt-4.0 surpasses the state of the art performance if the task is constrained to a subset of the target ontology where there is prior knowledge of the terms that are expected to be matched. While the results are promising, the non-deterministic nature of the outcomes, the high cost and the lack of concordance between different runs using the same prompt and input make the use of these LLMs challenging for this particular task.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2021

An Introduction to Word Embeddings and Language Models

Language models have advanced at a phenomenal pace over the past decade. This document provides a short introduction to terminology, word embeddings (aka low-dimensional representations), and popular large-scale language models (LMs). Word embeddings are used to represent words as numerical vectors and are context-independent, meaning a word can only have a single representation (e.g., club can only be club sandwich, not golf club ). Language models can determine the probability of a given sequence of words occurring in a sentence and can provide context to distinguish between words and phrases that sound similar. LMs are context-dependent (e.g., club can be club sandwich or golf club ) and largely fall in two main classes – autoregressive and autoencoding models. Autoregressive models are pretrained on the classic language modeling task: guess the next token having read all the previous ones. Those models can be fine-tuned and achieve great results on many tasks, the most natural application is text generation. A typical example of such models is GPT, but others include GPT-2, GPT-3, CTLR, TRANSFORMER-XL, REFORMER, XLNET. Autoencoding models are pretrained by corrupting the input tokens in some way and trying to reconstruct the original sentence. They can be fine-tuned and achieve great results on many tasks such as text generation, but their most natural application is sentence classification or token classification. A typical example of such models is BERT, but others include ROBERTA, ALBERT, XML, XML-ROBERTA, FLAUBERT AND LONGFORMER.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Transferring a Molecular Foundation Model for Polymer Property Predictions

Transformer-based large language models have remarkable potential to accelerate design optimization for applications such as drug development and material discovery. Self-supervised pretraining of transformer models requires large-scale data sets, which are often sparsely populated in topical areas such as polymer science. Further, state-of-the-art approaches for polymers conduct data augmentation to generate additional samples but unavoidably incur extra computational costs. In contrast, large-scale open-source data sets are available for small molecules and provide a potential solution to data scarcity through transfer learning. In this work, we show that using transformers pretrained on small molecules and fine-tuned on polymer properties achieves comparable accuracy to those trained on augmented polymer data sets for a series of benchmark prediction tasks.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

AGFormer: Adaptive Spatiotemporal graph informed transformer for multi-reservoir inflow forecasting

Accurate reservoir inflow forecasting is crucial for effective water resource management, yet most machine learning models focus on single-reservoir prediction and overlook spatial dependencies among hydrologically connected reservoirs. Here, we propose AGFormer (Adaptive Graph-Informed Transformer), an end-to-end framework that integrates adaptive graph learning with temporal sequence modeling for multi-reservoir inflow forecasting. A shared encoder and graph attention mechanism generate reservoir-specific embeddings, which are then processed by the Transformer-based encoder–decoder for multi-step inflow forecasting. We also introduce a pretraining paradigm to learn robust temporal embeddings from misaligned historical records. Evaluated on 30 reservoirs in the Upper Colorado River Basin, AGFormer achieves superior seven-day-ahead forecasts, with NSE > 0.75 for 20 reservoirs—outperforming Encoder–Decoder LSTM, GCN+LSTM, and Transformer baselines. Adaptive graph learning captures dynamic inter-reservoir dependencies, and feature attribution aligns with snowmelt-driven hydrology. Incorporating forecasted meteorological inputs further enhances accuracy, demonstrating AGFormer’s potential to support reservoir management under dynamic hydrological conditions.

Adaptive graph learning↗
DOE OSTI2026

SmileyLlama: modifying large language models for directed chemical space exploration

Here we show that large language models (LLMs) can be transformed via supervised fine-tuning of engineered prompts into SmileyLlama for exploring the chemical space of drug molecules. We benchmark SmileyLlama against pretrained LLMs and chemical language models trained from scratch for generating valid and novel drug-like molecules, and use direct preference optimization to both improve SmileyLlama’s adherence to a prompt and as part of the iMiner reinforcement learning framework to predict molecules with optimized three-dimensional conformations and high binding affinity to drug targets. By training an LLM to speak directly as a chemical language model, while retaining most of its natural language capabilities, we show that SmileyLlama can reliably generate molecules with user-specified properties rather than acting only as a chatbot with knowledge of chemistry or as a virtual assistant. While SmileyLlama is geared toward drug discovery, the supervised fine-tuning/direct preference optimization/LLM framework can be extended to other chemical, biological and materials applications.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

A knowledge-informed large language model framework for U.S. nuclear power plant shutdown initiating event classification for probabilistic risk assessment

Identifying and classifying shutdown initiating events (SDIEs) is critical for developing shutdown probabilistic risk assessment for nuclear power plants. Existing computational approaches cannot achieve satisfactory performance due to the challenges of unavailable large, labeled datasets, imbalanced event types, and label noise. To address these challenges, we propose a hybrid pipeline that integrates a knowledge-informed machine learning model to prescreen non-SDIEs and a large language model (LLM) to classify SDIEs into four types. In the prescreening stage, we proposed a set of 44 SDIE text patterns that consist of the most salient keywords and phrases from six SDIE types. Text vectorization based on the SDIE patterns generates feature vectors that are highly separable by using a simple binary classifier. The second stage builds Bidirectional Encoder Representations from Transformers (BERT)-based LLM, which learns generic English language representations from self-supervised pretraining on a large dataset and adapts to SDIE classification by fine-tuning it on an SDIE dataset. The proposed approaches are evaluated on a dataset with 10,928 events using precision, recall ratio, F 1 score, and average accuracy. In conclusion, the results demonstrate that the prescreening stage can exclude more than 97% non-SDIEs, and the LLM achieves an average accuracy of 95.1% for SDIE classification.

99 - GENERAL AND MISCELLANEOUS↗
DOE OSTI2026

Emerging Flexible Designs for Geospatial Multimodal Foundation Models

Foundation models are rapidly transforming Earth observation by enabling scalable pretraining across diverse unlabeled geospatial modalities. However, their architectural diversity—ranging from encoder-only to encoder-decoder and masked autoencoding paradigms—makes it challenging to assess performance trade-offs in a consistent manner. In this work, we present an apples-to-apples comparison of leading FM architectures designed for geospatial multimodal reasoning, with a particular focus on flexibility across varied spectral band configurations. We standardize pretraining using identical self-supervised learning objectives and training datasets, and evaluate all models under consistent parameterization on the GEOBench benchmark across classification and segmentation tasks. Our results offer new insights into the design trade-offs between model flexibility, modality alignment, and downstream task performance. By highlighting architectural strengths and limitations under controlled conditions, this study provides practical guidance for building next-generation geospatial foundation models capable of robust multimodal reasoning.

Ambrozio Dias, Philipe [ORNL] (ORCID:0000000194277↗
DOE OSTI2020

Deep Learning for Intelligent Bubble Size Detection in the Spallation Neutron Source Visual Target

The Spallation Neutron Source (SNS) at Oak Ridge National Laboratory (ORNL) will undergo proton power upgrade (PPU), increasing the proton beam power from 1.4 MW to 2.8 MW. From 2.8 MW, 2.0 MW will go to the current First Target Station and the rest will go to the future Second Target Station (STS). The First Target Station uses a liquid mercury target that is contained in a 316L stainless steel vessel. The proton beam is pulsed at 60 Hz, with a pulse of about 0.7μs. When the proton beam hits the target, the intense energy deposition leads to a rapid rise in temperature in the mercury. This temperature rise creates pressure waves that propagate through the mercury and cause cavitation erosion. The power upgrade will cause stronger pressure waves that will further increase damage because of cavitation. Injecting small helium bubbles in the mercury has been an efficient method of mitigating the pressure wave at 1.4 MW. However, at higher power, additional mitigation is necessary. Therefore, the 2 MW target vessel will be equipped with swirl bubblers and an additional gas injection port near the nose to inject more gas in the target. To develop a gas injection strategy and design, flow visualization in water with a transparent prototypical target (“visual target”) was performed. Bubble sizes and their spatial distribution in the flow loop are crucial to understanding the effectiveness of the bubbles in mitigating pressure waves. Bubbles were generated in the visual target under varied conditions of input pressures with helium and air. Images were captured using a high-speed camera at varied frame rates at different positions away from the swirl bubbler and different depths in the flow loop under varying lighting conditions. Initially, methods such as circular Hough transforms were applied after a series of images processing to obtain a general distribution of bubble sizes. Bubbles smaller than 500 μm are preferred to effectively mitigate the effect of pressure waves, which demands an accurate bubble detection and sizing system. Intelligent detection and identification of bubble sizes alleviate misdetection and improves accuracies. Employing neural networks, intelligent detection of bubble sizes and their distribution was developed and provides a robust alternative to traditional techniques. Human intervention was employed to label in-focus and out-of-focus bubbles in the set of training images. An object detection network using a pretrained convolutional neural network was created that extracted the features from the training images. Data augmentation was used to improve network accuracy through a random transformation of the original data.

Rasheed, Fayaz↗
DOE OSTI2025

DeepAndes: A Self-Supervised Vision Foundation Model for Multispectral Remote Sensing Imagery of the Andes

By mapping sites at large scales usingremotely sensed data, archaeologists can generate unique insights into long-term demographic trends, interregional social networks, and human adaptations in the past. Remote sensing surveys complement field-based approaches, and their reach can be especially great when combined with deep learning and computer vision techniques. However, conventional supervised deep learning methods face challenges in annotating fine-grained archaeological features at scale. In addition, while recent vision foundation models have shown remarkable success in learning large-scale remote sensing data with minimal annotations, most off-the-shelf solutions are designed for RGB images rather than multispectral satellite imagery, such as the eight-band data used in our study. In this article, we introduce DeepAndes, a transformer-based vision foundation model trained on three million multispectral satellite images, specifically tailored for Andean archaeology. DeepAndes incorporates a customized DINOv2 self-supervised learning algorithm optimized for eight-band multispectral imagery, marking the first foundation model designed explicitly for the Andes region. We evaluate its image understanding performance through imbalanced image classification, image instance retrieval, and pixel-level semantic segmentation tasks. Our experiments show that DeepAndes achieves superior F1 scores, mean average precision, and Dice scores in few-shot learning scenarios, significantly outperforming models trained from scratch or pretrained on smaller datasets. This underscores the effectiveness of large-scale self-supervised pretraining in archaeological remote sensing.

Guo, Junlin [Vanderbilt Univ., Nashville, TN (Unit↗