Jeffrey J. Ma

Optimizing the performance of large-scale software repositories demands expertise in code reasoning and software engineering (SWE) to reduce runtime while preserving program correctness. However, most benchmarks emphasize what to fix rather than how to fix code. We introduce SWE-fficiency, a benchmark for evaluating repository-level performance optimization on real workloads. Our suite contains 498 tasks across nine widely used data-science, machine-learning, and HPC repositories (e.g., numpy, pandas, scipy): given a complete codebase and a slow workload, an agent must investigate code semantics, localize bottlenecks and relevant tests, and produce a patch that matches or exceeds expert speedup while passing the same unit tests. To enable this how-to-fix evaluation, our automated pipeline scrapes GitHub pull requests for performance-improving edits, combining keyword filtering, static analysis, coverage tooling, and execution validation to both confirm expert speedup baselines and identify relevant repository unit tests. Empirical evaluation of state-of-the-art agents reveals significant underperformance. On average, agents achieve less than 0.15x the expert speedup: agents struggle in localizing optimization opportunities, reasoning about execution across functions, and maintaining correctness in proposed edits. We release the benchmark and accompanying data pipeline to facilitate research on automated performance engineering and long-horizon software reasoning.

The field of computer architecture, which bridges high-level software abstractions and low-level hardware implementations, remains absent from current large language model (LLM) evaluations. To this end, we present QuArch (pronounced ’quark’), the first benchmark designed to facilitate the development and evaluation of LLM knowledge and reasoning capabilities specifically in computer architecture. QuArch provides a comprehensive collection of 2,671 expert-validated question-answer (QA) pairs covering various aspects of computer architecture, including processor design, memory systems, and interconnection networks. Our evaluation reveals that while frontier models possess domain-specific knowledge, they struggle with skills that require higher-order thinking in computer architecture. Frontier model accuracies vary widely (from 34% to 72%) on these advanced questions, highlighting persistent gaps in architectural reasoning across analysis, design, and implementation QAs. By holistically assessing fundamental skills, QuArch provides a foundation for building and measuring LLM capabilities that can accelerate innovation in computing systems. With over 140 contributors from 40 institutions, this benchmark represents a community effort to set the standard for architectural reasoning in LLM evaluation.

As the scale of training large language models (LLMs) increases, one emergent failure is silent data corruption (SDC), where hardware produces incorrect computations without explicit failure signals. In this work, we are the first to investigate the impact of real-world SDCs on LLM training by comparing model training between healthy production nodes and unhealthy nodes exhibiting SDCs. With the help from a cloud computing platform, we access the unhealthy nodes that were swept out from production by automated fleet management. Using deterministic execution via XLA compiler and our proposed synchronization mechanisms, we isolate and analyze the impact of SDC errors on these nodes at three levels: at each submodule computation, at a single optimizer step, and at a training period. Our results reveal that the impact of SDCs on computation varies on different unhealthy nodes. Although in most cases the perturbations from SDCs on submodule computation and gradients are relatively small, SDCs can lead models to converge to different optima with different weights and even cause spikes in the training loss. Our analysis sheds light on further understanding and mitigating the impact of SDCs.

As the scale of training large language models (LLMs) increases, one emergent failure is silent data corruption (SDC), where hardware produces incorrect computations without explicit failure signals. In this work, we summarize the first investigation of the impact of real-world SDCs on LLM training presented in our ACL work [1]. In our investigation, we compare model training between healthy production nodes and unhealthy nodes exhibiting SDCs. With the help from a cloud computing platform, we access the unhealthy nodes that were swept out from production by automated fleet management. Using deterministic execution via XLA compiler and our proposed synchronization mechanisms, we isolate and analyze the impact of SDC errors on these nodes at three levels: at each submodule computation, at a single optimizer step, and at a training period. Our results reveal that the impact of SDCs on computation varies on different unhealthy nodes. Although in most cases the perturbations from SDCs on submodule computation and gradients are relatively small, SDCs can lead models to converge to different optima with different weights and even cause spikes in the training loss.

Federated Learning (FL) endeavors to harness decentralized data while preserving privacy, facing challenges of performance, scalability, and collaboration. Asynchronous Federated Learning (AFL) methods have emerged as promising alternatives to their synchronous counterparts bounded by the slowest agent, yet they add additional challenges in convergence guarantees, fairness with respect to compute heterogeneity, and incorporation of staleness in aggregated updates. Specifically, AFL biases model training heavily towards agents who can produce updates faster, leaving slower agents behind, who often also have differently distributed data which is not learned by the global model. Naively upweighting introduces incentive issues, where true fast updating agents may falsely report updates at a slower speed to increase their contribution to model training. We introduce FedStaleWeight, an algorithm addressing fairness in aggregating asynchronous client updates by employing average staleness to compute fair re-weightings. FedStaleWeight reframes asynchronous federated learning aggregation as a mechanism design problem, devising a weighting strategy that incentivizes truthful compute speed reporting without favoring faster update-producing agents by upweighting agent updates based on staleness. Leveraging only observed agent update staleness, FedStaleWeight results in more equitable aggregation on a per-agent basis. We both provide theoretical convergence guarantees in the smooth, non-convex setting and empirically compare FedStaleWeight against the commonly used asynchronous FedBuff with gradient averaging, demonstrating how it achieves stronger fairness, expediting convergence to a higher global model accuracy. Finally, we provide an open-source test bench to facilitate exploration of buffered AFL aggregation strategies, fostering further research in asynchronous federated learning paradigms.

Many economic games and machine learning approaches can be cast as competitive optimization problems where multiple agents are minimizing their respective objective function, which depends on all agents’ actions. While gradient descent is a reliable basic workhorse for single-agent optimization, it often leads to oscillation in competitive optimization. In this work we propose polymatrix competitive gradient descent (PCGD) as a method for solving general sum competitive optimization involving arbitrary numbers of agents. The updates of our method are obtained as the Nash equilibria of a local polymatrix approximation with a quadratic regularization, and can be computed efficiently by solving a linear system of equations. We prove local convergence of PCGD to stable fixed points for n-player general-sum games, and show that it does not require adapting the step size to the strength of the player-interactions. We use PCGD to optimize policies in multi-agent reinforcement learning and demonstrate its advantages in Snake, Markov soccer and an electricity market game. Agents trained by PCGD outperform agents trained with simultaneous gradient descent, symplectic gradient adjustment, and extragradient in Snake and Markov soccer games and on the electricity market game, PCGD trains faster than both simultaneous gradient descent and the extragradient method.