Neural network-based emulators for the inference of stellar parameters and elemental abundances represent an increasingly popular methodology in modern spectroscopic surveys. However, these approaches are often constrained by their emulation precision and domain transfer capabilities. Greater generalizability has previously been achieved only with significantly larger model architectures, as demonstrated by Transformer-based models in natural language processing. This observation aligns with neural scaling laws, where model performance predictably improves with increased model size, computational resources allocated to model training, and training data volume. In this study, we demonstrate that these scaling laws also apply to Transformer-based spectral emulators in astronomy. Building upon our previous work with TransformerPayne and incorporating Maximum Update Parametrization techniques from natural language models, we provide training guidelines for scaling models to achieve optimal performance. Our results show that within the explored parameter space, clear scaling relationships emerge. These findings suggest that optimal computational resource allocation requires balanced scaling. Specifically, given a tenfold increase in training compute, achieving an optimal seven-fold reduction in mean squared error necessitates an approximately 2.5-fold increase in dataset size and a 3.8-fold increase in model size. This study establishes a foundation for developing spectral foundational models with enhanced domain transfer capabilities.

Recent advancements have positioned AI, and particularly Large Language Models (LLMs), as transformative tools for scientific research, capable of addressing complex tasks that require reasoning, problem-solving, and decision-making. Their exceptional capabilities suggest their potential as scientific research assistants but also highlight the need for holistic, rigorous, and domain-specific evaluation to assess effectiveness in real-world scientific applications. This paper describes a multifaceted methodology for Evaluating AI models as scientific Research Assistants (EAIRA) developed at Argonne National Laboratory. This methodology incorporates four primary classes of evaluations. 1) Multiple Choice Questions to assess factual recall; 2) Open Response to evaluate advanced reasoning and problem-solving skills; 3) Lab-Style Experiments involving detailed analysis of capabilities as research assistants in controlled environments; and 4) Field-Style Experiments to capture researcher-LLM interactions at scale in a wide range of scientific domains and applications. These complementary methods enable a comprehensive analysis of LLM strengths and weaknesses with respect to their scientific knowledge, reasoning abilities, and adaptability. Recognizing the rapid pace of LLM advancements, we designed the methodology to evolve and adapt so as to ensure its continued relevance and applicability. This paper describes the methodology state at the end of February 2025. Although developed within a subset of scientific domains, the methodology is designed to be generalizable to a wide range of scientific domains.

Obtaining well-calibrated photometric redshift probability densities for galaxies without a spectroscopic measurement remains a challenge. Deep learning discriminative models, typically fed with multi-band galaxy images, can produce outputs that mimic probability densities and achieve state-of-the-art accuracy. However, such models may be affected by miscalibration that would result in discrepancies between the model outputs and the actual distributions of true redshifts. Our work develops a novel method called the Contrastive Learning and Adaptive KNN for Photometric Redshift (CLAP) that resolves this issue. It leverages supervised contrastive learning (SCL) and k-nearest neighbours (KNN) to construct and calibrate raw probability density estimates, and implements a refitting procedure to resume end-to-end discriminative models ready to produce final estimates for large-scale imaging data. The harmonic mean is adopted to combine an ensemble of estimates from multiple realisations for improving accuracy. Our experiments demonstrate that CLAP takes advantage of both deep learning and KNN, outperforming benchmark methods on the calibration of probability density estimates and retaining high accuracy and computational efficiency. With reference to CLAP, we point out that miscalibration is particularly sensitive to the method-induced excessive correlations among data instances in addition to the unaccounted-for epistemic uncertainties. Reducing the uncertainties may not guarantee the removal of miscalibration due to the presence of such excessive correlations, yet this is a problem for conventional deep learning methods rather than CLAP. These discussions underscore the robustness of CLAP for obtaining photometric redshift probability densities required by astrophysical and cosmological applications. This is the first paper in our series on CLAP.

Continual pretraining of large language models on domain-specific data has been proposed to enhance performance on downstream tasks. In astronomy, the previous absence of astronomy-focused benchmarks has hindered objective evaluation of these specialized LLM models. Leveraging a recent initiative to curate high-quality astronomical MCQs, this study aims to quantitatively assess specialized LLMs in astronomy. We find that the previously released AstroLLaMA series, based on LLaMA-2-7B, underperforms compared to the base model. We demonstrate that this performance degradation can be partially mitigated by utilizing high-quality data for continual pretraining, such as summarized text from arXiv. Despite the observed catastrophic forgetting in smaller models, our results indicate that continual pretraining on the 70B model can yield significant improvements. However, the current supervised fine-tuning dataset still constrains the performance of instruct models. In conjunction with this study, we introduce a new set of models, AstroLLaMA-3-8B and AstroLLaMA-2-70B, building upon the previous AstroLLaMA series.

We present a comprehensive evaluation of proprietary and open-weights large language models using the first astronomy-specific benchmarking dataset. This dataset comprises 4,425 multiple-choice questions curated from the Annual Review of Astronomy and Astrophysics, covering a broad range of astrophysical topics. Our analysis examines model performance across various astronomical subfields and assesses response calibration, crucial for potential deployment in research environments. Claude-3.5-Sonnet outperforms competitors by up to 4.6 percentage points, achieving 85.0% accuracy. For proprietary models, we observed a universal reduction in cost every 3-to-12 months to achieve similar score in this particular astronomy benchmark. Open-source models have rapidly improved, with LLaMA-3-70b (80.6%) and Qwen-2-72b (77.7%) now competing with some of the best proprietary models. We identify performance variations across topics, with non-English-focused models generally struggling more in exoplanet-related fields, stellar astrophysics, and instrumentation related questions. These challenges likely stem from less abundant training data, limited historical context, and rapid recent developments in these areas. This pattern is observed across both open-weights and proprietary models, with regional dependencies evident, highlighting the impact of training data diversity on model performance in specialized scientific domains. Top-performing models demonstrate well-calibrated confidence, with correlations above 0.9 between confidence and correctness, though they tend to be slightly underconfident. The development for fast, low-cost inference of open-weights models presents new opportunities for affordable deployment in astronomy. The rapid progress observed suggests that LLM-driven research in astronomy may become feasible in the near future.

Analyzing time series of fluxes from stars, known as stellar light curves, can reveal valuable information about stellar properties. However, most current methods rely on extracting summary statistics, and studies using deep learning have been limited to supervised approaches. In this research, we investigate the scaling law properties that emerge when learning from astronomical time series data using self-supervised techniques. By employing the GPT-2 architecture, we show the learned representation improves as the number of parameters increases from $10^4$ to $10^9$, with no signs of performance plateauing. We demonstrate that a self-supervised Transformer model achieves 3-10 times the sample efficiency compared to the state-of-the-art supervised learning model when inferring the surface gravity of stars as a downstream task. Our research lays the groundwork for analyzing stellar light curves by examining them through large-scale auto-regressive generative models.

Stellar light curves contain valuable information about oscillations and granulation, offering insights into stars' internal structures and evolutionary states. Traditional asteroseismic techniques, primarily focused on power spectral analysis, often overlook the crucial phase information in these light curves. Addressing this gap, recent machine learning applications, particularly those using Convolutional Neural Networks (CNNs), have made strides in inferring stellar properties from light curves. However, CNNs are limited by their localized feature extraction capabilities. In response, we introduce $\textit{Astroconformer}$, a Transformer-based deep learning framework, specifically designed to capture long-range dependencies in stellar light curves. Our empirical analysis centers on estimating surface gravity ($\log g$), using a dataset derived from single-quarter Kepler light curves with $\log g$ values ranging from 0.2 to 4.4. $\textit{Astroconformer}$ demonstrates superior performance, achieving a root-mean-square-error (RMSE) of 0.017 dex at $\log g\approx3$ in data-rich regimes and up to 0.1 dex in sparser areas. This performance surpasses both K-nearest neighbor models and advanced CNNs. Ablation studies highlight the influence of receptive field size on model effectiveness, with larger fields correlating to improved results. $\textit{Astroconformer}$ also excels in extracting $\nu_{\max}$ with high precision. It achieves less than 2% relative median absolute error for 90-day red giant light curves. Notably, the error remains under 3% for 30-day light curves, whose oscillations are undetectable by a conventional pipeline in 30% cases. Furthermore, the attention mechanisms in $\textit{Astroconformer}$ align closely with the characteristics of stellar oscillations and granulation observed in light curves.

To enhance this, we introduce AstroLLaMA-Chat, an advanced version of AstroLLaMA. This new iteration broadens the training scope to include introductions and conclusions of papers, alongside abstracts, as these sections are often rich in pivotal information for question-answering tasks. We initiated by downloading all papers up to July 2023, including all the files that come with a submission to arXiv. The data has been further refined for optimal operability, retaining only files with ".tex" suffixes. Through a multi-stage process, and utilising a comprehensive regex matching process, the extraction of the targeted sections was performed. Given the diverse LaTeX formatting standards, approximately 90% of the samples remained post-processing. Subsequently, we removed specific formatting patterns, comments, and superfluous symbols like newlines to ensure the readability of the training data. Further, we have fine-tuned AstroLLaMA-Chat on a domain-specific dialogue dataset. To generate Question-Answer pairs, we engaged GPT-4 (OpenAI 2023) to formulate pertinent questions from paragraphs within 300,000 arXiv papers, with GPT-4 also tasked with answering these questions by retrieving context-relevant information.

Large language models excel in many human-language tasks but often falter in highly specialized domains like scholarly astronomy. To bridge this gap, we introduce AstroLLaMA, a 7-billion-parameter model fine-tuned from LLaMA-2 using over 300,000 astronomy abstracts from arXiv. Optimized for traditional causal language modeling, AstroLLaMA achieves a 30% lower perplexity than Llama-2, showing marked domain adaptation. Our model generates more insightful and scientifically relevant text completions and embedding extraction than state-of-the-arts foundation models despite having significantly fewer parameters. AstroLLaMA serves as a robust, domain-specific model with broad fine-tuning potential. Its public release aims to spur astronomy-focused research, including automatic paper summarization and conversational agent development.

ABSTRACT We demonstrate the potential of the state-of-the-art OpenAI GPT-4 large language model to engage in meaningful interactions with Astronomy papers using in-context prompting. To optimize for efficiency, we employ a distillation technique that effectively reduces the size of the original input paper by 50%, while maintaining the paragraph structure and overall semantic integrity. We then explore the model's responses using a multi-document context (ten distilled documents). Our findings indicate that GPT-4 excels in the multi-document domain, providing detailed answers contextualized within the framework of related research findings. INTRODUCTION Large language models (LLMs) have significantly advanced natural language processing, allowing machines to process and generate intricate text with remarkable quality (e.g., Devlin et al. 2018; Brown et al. 2020; Chowdhery et al. 2022; Bubeck et al. 2023).