Shang Zhu

Advika Vidhyadhiraja, Arun Pa Thiagarajan, Shang Zhu et al.

Learning exchange correlation functionals, used in quantum chemistry calculations, from data has become increasingly important in recent years, but training such a functional requires sophisticated software infrastructure. For this reason, we build open source infrastructure to train neural exchange correlation functionals. We aim to standardize the processing pipeline by adapting state-of-the-art techniques from work done by multiple groups. We have open sourced the model in the DeepChem library to provide a platform for additional research on differentiable quantum chemistry methods.

Junlin Wang, Federico Bianchi, Shang Zhu et al.

Modern AI benchmarks operate at a complexity that outpaces traditional verification methods. Tasks authored by domain experts often contain implicit assumptions, incomplete environment specifications, and brittle evaluation logic that human annotation cannot reliably catch. We introduce Auto Benchmark Audit (ABA), an agentic framework that systematically audits individual benchmark tasks, uncovering issues such as hidden environment dependencies, specification gaps, and limited grading logic. We run ABA on a collection of frontier LLM benchmarks and previous NeurIPS publications, totaling 168 benchmarks across nine domains. Across this corpus, ABA identifies critical issues including ambiguous task design, execution environment conflicts, and incorrect ground truths in over 25.7% of the evaluated tasks. The precision of these automated audits is validated by expert review and independent third-party reports such as upstream PRs. Crucially, we demonstrate that these problematic tasks severely distorts capability assessments for agents and LLMs: filtering out these tasks with issues shifts model rankings and increases average performance on SWE-bench Verified and Terminal-Bench 2 by 9.9% and 9.6%, respectively. We release the agentic tool and all task annotations to support the future development of frontier benchmarks.

Jon Saad-Falcon, Avanika Narayan, Hakki Orhun Akengin et al.

Large language model (LLM) queries are predominantly processed by frontier models in centralized cloud infrastructure. Rapidly growing demand strains this paradigm, and cloud providers struggle to scale infrastructure at pace. Two advances enable us to rethink this paradigm: small LMs (<=20B active parameters) now achieve competitive performance to frontier models on many tasks, and local accelerators (e.g., Apple M4 Max) run these models at interactive latencies. This raises the question: can local inference viably redistribute demand from centralized infrastructure? Answering this requires measuring whether local LMs can accurately answer real-world queries and whether they can do so efficiently enough to be practical on power-constrained devices (i.e., laptops). We propose intelligence per watt (IPW), task accuracy divided by unit of power, as a metric for assessing capability and efficiency of local inference across model-accelerator pairs. We conduct a large-scale empirical study across 20+ state-of-the-art local LMs, 8 accelerators, and a representative subset of LLM traffic: 1M real-world single-turn chat and reasoning queries. For each query, we measure accuracy, energy, latency, and power. Our analysis reveals $3$ findings. First, local LMs can accurately answer 88.7% of single-turn chat and reasoning queries with accuracy varying by domain. Second, from 2023-2025, IPW improved 5.3x and local query coverage rose from 23.2% to 71.3%. Third, local accelerators achieve at least 1.4x lower IPW than cloud accelerators running identical models, revealing significant headroom for optimization. These findings demonstrate that local inference can meaningfully redistribute demand from centralized infrastructure, with IPW serving as the critical metric for tracking this transition. We release our IPW profiling harness for systematic intelligence-per-watt benchmarking.

Shang Zhu, Bharath Ramsundar, Emil Annevelink et al.

Electrolytes play a critical role in designing next-generation battery systems, by allowing efficient ion transfer, preventing charge transfer, and stabilizing electrode-electrolyte interfaces. In this work, we develop a differentiable geometric deep learning (GDL) model for chemical mixtures, DiffMix, which is applied in guiding robotic experimentation and optimization towards fast-charging battery electrolytes. In particular, we extend mixture thermodynamic and transport laws by creating GDL-learnable physical coefficients. We evaluate our model with mixture thermodynamics and ion transport properties, where we show improved prediction accuracy and model robustness of DiffMix than its purely data-driven variants. Furthermore, with a robotic experimentation setup, Clio, we improve ionic conductivity of electrolytes by over 18.8% within 10 experimental steps, via differentiable optimization built on DiffMix gradients. By combining GDL, mixture physics laws, and robotic experimentation, DiffMix expands the predictive modeling methods for chemical mixtures and enables efficient optimization in large chemical spaces.

Yoel Zimmermann, Adib Bazgir, Alexander Al-Feghali et al.

Large Language Models (LLMs) are reshaping many aspects of materials science and chemistry research, enabling advances in molecular property prediction, materials design, scientific automation, knowledge extraction, and more. Recent developments demonstrate that the latest class of models are able to integrate structured and unstructured data, assist in hypothesis generation, and streamline research workflows. To explore the frontier of LLM capabilities across the research lifecycle, we review applications of LLMs through 34 total projects developed during the second annual Large Language Model Hackathon for Applications in Materials Science and Chemistry, a global hybrid event. These projects spanned seven key research areas: (1) molecular and material property prediction, (2) molecular and material design, (3) automation and novel interfaces, (4) scientific communication and education, (5) research data management and automation, (6) hypothesis generation and evaluation, and (7) knowledge extraction and reasoning from the scientific literature. Collectively, these applications illustrate how LLMs serve as versatile predictive models, platforms for rapid prototyping of domain-specific tools, and much more. In particular, improvements in both open source and proprietary LLM performance through the addition of reasoning, additional training data, and new techniques have expanded effectiveness, particularly in low-data environments and interdisciplinary research. As LLMs continue to improve, their integration into scientific workflows presents both new opportunities and new challenges, requiring ongoing exploration, continued refinement, and further research to address reliability, interpretability, and reproducibility.

Junlin Wang, Roy Xie, Shang Zhu et al.

Building helpful and harmless large language models (LLMs) requires effective model alignment approach based on human instructions and feedback, which necessitates high-quality human-labeled data. Constructing such datasets is often expensive and hard to scale, and may face potential limitations on diversity and generalization. To address these challenges, we introduce Mixture of Agents Alignment (MoAA), that leverages the collective strengths of various language models to provide high-quality data for model alignment. By employing MoAA, we enhance both supervised fine-tuning and preference optimization, leading to improved performance compared to using a single model alone to generate alignment data (e.g. using GPT-4o alone). Evaluation results show that our approach can improve win rate of LLaMA-3.1-8B-Instruct from 19.5 to 48.3 on Arena-Hard and from 22.33 to 57.23 on AlpacaEval2, highlighting a promising direction for model alignment through this new scalable and diverse synthetic data recipe. Furthermore, we demonstrate that MoAA enables a self-improvement pipeline, where models finetuned on MoA-generated data surpass their own initial capabilities, providing evidence that our approach can push the frontier of open-source LLMs without reliance on stronger external supervision. Data and code will be released.

Fan Nie, Junlin Wang, Harper Hua et al.

Data science agents promise to accelerate discovery and insight-generation by turning data into executable analyses and findings. Yet existing data science benchmarks fall short due to fragmented evaluation interfaces that make cross-benchmark comparison difficult, narrow task coverage and a lack of rigorous data grounding. In particular, we show that a substantial portion of tasks in current benchmarks can be solved without using the actual data. To address these limitations, we introduce DSGym, a standardized framework for evaluating and training data science agents in self-contained execution environments. Unlike static benchmarks, DSGym provides a modular architecture that makes it easy to add tasks, agent scaffolds, and tools, positioning it as a live, extensible testbed. We curate DSGym-Tasks, a holistic task suite that standardizes and refines existing benchmarks via quality and shortcut solvability filtering. We further expand coverage with (1) DSBio: expert-derived bioinformatics tasks grounded in literature and (2) DSPredict: challenging prediction tasks spanning domains such as computer vision, molecular prediction, and single-cell perturbation. Beyond evaluation, DSGym enables agent training via execution-verified data synthesis pipeline. As a case study, we build a 2,000-example training set and trained a 4B model in DSGym that outperforms GPT-4o on standardized analysis benchmarks. Overall, DSGym enables rigorous end-to-end measurement of whether agents can plan, implement, and validate data analyses in realistic scientific context.

Linda He, Jue Wang, Maurice Weber et al.

Large Language Models (LLMs) struggle with long-context reasoning, not only due to the quadratic scaling of computational complexity with sequence length but also because of the scarcity and expense of annotating long-context data. There has been barely any open-source work that systematically ablates long-context data, nor is there any openly available instruction tuning dataset with contexts surpassing 100K tokens. To bridge this gap, we introduce a novel post-training synthetic data generation strategy designed to efficiently extend the context window of LLMs while preserving their general task performance. Our approach scalably extends to arbitrarily long context lengths, unconstrained by the length of available real-world data, which effectively addresses the scarcity of raw long-context data. Through a step-by-step rotary position embedding (RoPE) scaling training strategy, we demonstrate that our model, with a context length of up to 1M tokens, performs well on the RULER benchmark and InfiniteBench and maintains robust performance on general language tasks.

Yongchan Kwon, Shang Zhu, Federico Bianchi et al.

The ability of large language models (LLMs) to follow user instructions is central to their reliability, safety, and usefulness. While prior studies assess instruction adherence in the model's main responses, we argue that it is also critical for large reasoning models (LRMs) to follow user instructions throughout their reasoning process. Reasoning instruction following makes LRMs more controllable and transparent, while reducing risks of undesirable shortcuts, hallucinations, or reward hacking within reasoning traces. To evaluate this dimension, we introduce ReasonIF, a systematic benchmark for assessing reasoning instruction following. ReasonIF includes six categories of instruction prompts, spanning multilingual reasoning, formatting and length control. Across many open-source LRMs including GPT-OSS, Qwen3, and DeepSeek-R1, we find substantial failures in reasoning instruction adherence: the highest instruction following score (IFS) remains below 0.25, meaning that fewer than $25\%$ of reasoning traces comply with the given instructions. Notably, as task difficulty increases, reasoning instruction following degrades further. We also explore two strategies to enhance reasoning instruction fidelity. (1) multi-turn reasoning and (2) Reasoning Instruction Finetuning (RIF) using synthetic data. RIF improves the IFS of $GPT-OSS-20B$ from 0.11 to 0.27, indicating measurable progress but leaving ample room for improvement.

Shai Pranesh, Shang Zhu, Venkat Viswanathan et al.

Finding accurate solutions to the electronic Schrödinger equation plays an important role in discovering important molecular and material energies and characteristics. Consequently, solving systems with large numbers of electrons has become increasingly important. Variational Monte Carlo (VMC) methods, especially those approximated through deep neural networks, are promising in this regard. In this paper, we aim to integrate one such model called the FermiNet, a post-Hartree-Fock (HF) Deep Neural Network (DNN) model, into a standard and widely used open source library, DeepChem. We also propose novel initialization techniques to overcome the difficulties associated with the assignment of excess or lack of electrons for ions.

Junlin Wang, Shang Zhu, Jon Saad-Falcon et al.

There is intense interest in investigating how inference time compute (ITC) (e.g. repeated sampling, refinements, etc) can improve large language model (LLM) capabilities. At the same time, recent breakthroughs in reasoning models, such as Deepseek-R1, unlock the opportunity for reinforcement learning to improve LLM reasoning skills. An in-depth understanding of how ITC interacts with reasoning across different models could provide important guidance on how to further advance the LLM frontier. This work conducts a comprehensive analysis of inference-time scaling methods for both reasoning and non-reasoning models on challenging reasoning tasks. Specifically, we focus our research on verifier-free inference time-scaling methods due to its generalizability without needing a reward model. We construct the Pareto frontier of quality and efficiency. We find that non-reasoning models, even with an extremely high inference budget, still fall substantially behind reasoning models. For reasoning models, majority voting proves to be a robust inference strategy, generally competitive or outperforming other more sophisticated ITC methods like best-of-N and sequential revisions, while the additional inference compute offers minimal improvements. We further perform in-depth analyses of the association of key response features (length and linguistic markers) with response quality, with which we can improve the existing ITC methods. We find that correct responses from reasoning models are typically shorter and have fewer hedging and thinking markers (but more discourse markers) than the incorrect responses.

Yoel Zimmermann, Adib Bazgir, Zartashia Afzal et al.

Here, we present the outcomes from the second Large Language Model (LLM) Hackathon for Applications in Materials Science and Chemistry, which engaged participants across global hybrid locations, resulting in 34 team submissions. The submissions spanned seven key application areas and demonstrated the diverse utility of LLMs for applications in (1) molecular and material property prediction; (2) molecular and material design; (3) automation and novel interfaces; (4) scientific communication and education; (5) research data management and automation; (6) hypothesis generation and evaluation; and (7) knowledge extraction and reasoning from scientific literature. Each team submission is presented in a summary table with links to the code and as brief papers in the appendix. Beyond team results, we discuss the hackathon event and its hybrid format, which included physical hubs in Toronto, Montreal, San Francisco, Berlin, Lausanne, and Tokyo, alongside a global online hub to enable local and virtual collaboration. Overall, the event highlighted significant improvements in LLM capabilities since the previous year's hackathon, suggesting continued expansion of LLMs for applications in materials science and chemistry research. These outcomes demonstrate the dual utility of LLMs as both multipurpose models for diverse machine learning tasks and platforms for rapid prototyping custom applications in scientific research.

Ziqi Wang, Hongshuo Huang, Hancheng Zhao et al.

Materials discovery relies on high-throughput, high-fidelity simulation techniques such as Density Functional Theory (DFT), which require years of training, extensive parameter fine-tuning and systematic error handling. To address these challenges, we introduce the DFT-based Research Engine for Agentic Materials Screening (DREAMS), a hierarchical, multi-agent framework for DFT simulation that combines a central Large Language Model (LLM) planner agent with domain-specific LLM agents for atomistic structure generation, systematic DFT convergence testing, High-Performance Computing (HPC) scheduling, and error handling. In addition, a shared canvas helps the LLM agents to structure their discussions, preserve context and prevent hallucination. We validate DREAMS capabilities on the Sol27LC lattice-constant benchmark, achieving average errors below 1\% compared to the results of human DFT experts. Furthermore, we apply DREAMS to the long-standing CO/Pt(111) adsorption puzzle, demonstrating its long-term and complex problem-solving capabilities. The framework again reproduces expert-level literature adsorption-energy differences. Finally, DREAMS is employed to quantify functional-driven uncertainties with Bayesian ensemble sampling, confirming the Face Centered Cubic (FCC)-site preference at the Generalized Gradient Approximation (GGA) DFT level. In conclusion, DREAMS approaches L3-level automation - autonomous exploration of a defined design space - and significantly reduces the reliance on human expertise and intervention, offering a scalable path toward democratized, high-throughput, high-fidelity computational materials discovery.

Jon Saad-Falcon, E. Kelly Buchanan, Mayee F. Chen et al.

Verifiers can improve language model capabilities by scoring and ranking responses from generated candidates. Currently, high-quality verifiers are either unscalable (e.g., humans) or limited in utility (e.g., tools like Lean). While LM judges and reward models have become broadly useful as general-purpose verifiers, a significant performance gap remains between them and oracle verifiers (verifiers with perfect accuracy). To help close this gap, we introduce Weaver, a framework for designing a strong verifier by combining multiple weak, imperfect verifiers. We find weighted ensembles of verifiers, which typically require learning from labeled data, significantly outperform unweighted combinations due to differences in verifier accuracies. To reduce dependency on labeled data, Weaver leverages weak supervision to estimate each verifier's accuracy and combines outputs into a unified score that better reflects true response quality. However, directly applying weak supervision algorithms poses challenges, including inconsistent verifier output formats and handling low-quality verifiers. Weaver addresses these using dataset statistics to normalize outputs and filter specific verifiers. We study Weaver's effectiveness in test-time repeated sampling, where a model generates multiple candidate responses and selects one. Our evaluations show Weaver significantly improves over Pass@1-performance when selecting the first candidate-across reasoning and math tasks, achieving o3-mini-level accuracy with Llama 3.3 70B Instruct as generator, and an ensemble of 70B or smaller judge and reward models as verifiers (87.7% average). This gain mirrors the jump between GPT-4o and o3-mini (69.0% vs. 86.7%), which required extensive finetuning and post-training. To reduce computational costs of verifier ensembles, we train a 400M cross-encoder using Weaver's combined output scores.

Zhen Xu, Shang Zhu, Jue Wang et al.

We investigate the challenge of applying Large Language Models (LLMs) to long texts. We propose a theoretical framework that distinguishes the failure modes of long context tasks into three categories: cross-chunk dependence (task noise), confusion that grows with context size (model noise), and the imperfect integration of partial results (aggregator noise). Under this view, we analyze when it is effective to use multi-agent chunking, i.e., dividing a length sequence into smaller chunks and aggregating the processed results of each chunk. Our experiments on tasks such as retrieval, question answering, and summarization confirm both the theoretical analysis and the conditions that favor multi-agent chunking. By exploring superlinear model noise growth with input length, we also explain why, for large inputs, a weaker model configured with chunk-based processing can surpass a more advanced model like GPT4o applied in a single shot. Overall, we present a principled understanding framework and our results highlight a direct pathway to handling long contexts in LLMs with carefully managed chunking and aggregator strategies.

Changwen Xu, Shang Zhu, Venkatasubramanian Viswanathan

The prediction of crystal properties is essential for understanding structure-property relationships and accelerating the discovery of functional materials. However, conventional approaches relying on experimental measurements or density functional theory (DFT) calculations are often resource-intensive, limiting their scalability. Machine learning (ML) models offer a promising alternative by learning complex structure-property relationships from data, enabling faster predictions. Yet, existing ML models often rely on labeled data, adopt representations that poorly capture essential structural characteristics, and lack integration with physical principles--factors that limit their generalizability and interpretability. Here, we introduce CLOUD (Crystal Language mOdel for Unified and Differentiable materials modeling), a transformer-based framework trained on a novel Symmetry-Consistent Ordered Parameter Encoding (SCOPE) that encodes crystal symmetry, Wyckoff positions, and composition in a compact, coordinate-free string representation. Pre-trained on over six million crystal structures, CLOUD is fine-tuned on multiple downstream tasks and achieves competitive performance in predicting a wide range of material properties, demonstrating strong scaling performance. Furthermore, as proof of concept of differentiable materials modeling, CLOUD is applied to predict the phonon internal energy and heat capacity, which integrates the Debye model to preserve thermodynamic consistency. The CLOUD-DEBYE framework enforces thermodynamic consistency and enables temperature-dependent property prediction without requiring additional data. These results demonstrate the potential of CLOUD as a scalable and physics-informed foundation model for crystalline materials, unifying symmetry-consistent representations with physically grounded learning for property prediction and materials discovery.

Emil Annevelink, Rachel Kurchin, Eric Muckley et al.

Large-scale electrification is vital to addressing the climate crisis, but several scientific and technological challenges remain to fully electrify both the chemical industry and transportation. In both of these areas, new electrochemical materials will be critical, but their development currently relies heavily on human-time-intensive experimental trial and error and computationally expensive first-principles, meso-scale and continuum simulations. We present an automated workflow, AutoMat, that accelerates these computational steps by introducing both automated input generation and management of simulations across scales from first principles to continuum device modeling. Furthermore, we show how to seamlessly integrate multi-fidelity predictions such as machine learning surrogates or automated robotic experiments "in-the-loop". The automated framework is implemented with design space search techniques to dramatically accelerate the overall materials discovery pipeline by implicitly learning design features that optimize device performance across several metrics. We discuss the benefits of AutoMat using examples in electrocatalysis and energy storage and highlight lessons learned.