Spandan Garg

Spandan Garg, Roshanak Zilouchian Moghaddam, Colin B. Clement et al. · baidu, microsoft-research

Improving software performance is an important yet challenging part of the software development cycle. Today, the majority of performance inefficiencies are identified and patched by performance experts. Recent advancements in deep learning approaches and the wide-spread availability of open source data creates a great opportunity to automate the identification and patching of performance problems. In this paper, we present DeepPERF, a transformer-based approach to suggest performance improvements for C# applications. We pretrain DeepPERF on English and Source code corpora and followed by finetuning for the task of generating performance improvement patches for C# applications. Our evaluation shows that our model can generate the same performance improvement suggestion as the developer fix in ~53% of the cases, getting ~34% of them verbatim in our expert-verified dataset of performance changes made by C# developers. Additionally, we evaluate DeepPERF on 50 open source C# repositories on GitHub using both benchmark and unit tests and find that our model is able to suggest valid performance improvements that can improve both CPU usage and Memory allocations. So far we've submitted 19 pull-requests with 28 different performance optimizations and 11 of these PRs have been approved by the project owners.

Spandan Garg, Roshanak Zilouchian Moghaddam, Neel Sundaresan · microsoft-research

Performance bugs are non-functional bugs that can even manifest in well-tested commercial products. Fixing these performance bugs is an important yet challenging problem. In this work, we address this challenge and present a new approach called Retrieval-Augmented Prompt Generation (RAPGen). Given a code snippet with a performance issue, RAPGen first retrieves a prompt instruction from a pre-constructed knowledge-base of previous performance bug fixes and then generates a prompt using the retrieved instruction. It then uses this prompt on a Large Language Model (such as Codex) in zero-shot to generate a fix. We compare our approach with the various prompt variations and state of the art methods in the task of performance bug fixing. Our evaluation shows that RAPGen can generate performance improvement suggestions equivalent or better than a developer in ~60% of the cases, getting ~42% of them verbatim, in an expert-verified dataset of past performance changes made by C# developers.

Roshanak Zilouchian Moghaddam, Spandan Garg, Colin B. Clement et al. · microsoft-research

Continuous evolution in modern software often causes documentation, tutorials, and examples to be out of sync with changing interfaces and frameworks. Relying on outdated documentation and examples can lead programs to fail or be less efficient or even less secure. In response, programmers need to regularly turn to other resources on the web such as StackOverflow for examples to guide them in writing software. We recognize that this inconvenient, error-prone, and expensive process can be improved by using machine learning applied to software usage data. In this paper, we present our practical system which uses machine learning on large-scale telemetry data and documentation corpora, generating appropriate and complex examples that can be used to improve documentation. We discuss both feature-based and transformer-based machine learning approaches and demonstrate that our system achieves 100% coverage for the used functionalities in the product, providing up-to-date examples upon every release and reduces the numbers of PRs submitted by software owners writing and editing documentation by >68%. We also share valuable lessons learnt during the 3 years that our production quality system has been deployed for Azure Cloud Command Line Interface (Azure CLI).

Dhruv Gautam, Spandan Garg, Jinu Jang et al. · microsoft-research

Recent advances in language model (LM) agents and function calling have enabled autonomous, feedback-driven systems to solve problems across various digital domains. To better understand the unique limitations of LM agents, we introduce RefactorBench, a benchmark consisting of 100 large handcrafted multi-file refactoring tasks in popular open-source repositories. Solving tasks within RefactorBench requires thorough exploration of dependencies across multiple files and strong adherence to relevant instructions. Every task is defined by 3 natural language instructions of varying specificity and is mutually exclusive, allowing for the creation of longer combined tasks on the same repository. Baselines on RefactorBench reveal that current LM agents struggle with simple compositional tasks, solving only 22% of tasks with base instructions, in contrast to a human developer with short time constraints solving 87%. Through trajectory analysis, we identify various unique failure modes of LM agents, and further explore the failure mode of tracking past actions. By adapting a baseline agent to condition on representations of state, we achieve a 43.9% improvement in solving RefactorBench tasks. We further extend our state-aware approach to encompass entire digital environments and outline potential directions for future research. RefactorBench aims to support the study of LM agents by providing a set of real-world, multi-hop tasks within the realm of code.

Spandan Garg, Yufan Huang

While significant progress has been made in automating various aspects of software development through coding agents, there is still significant room for improvement in their bug fixing capabilities. Debugging and investigation of runtime behavior remains largely a manual, developer-driven process. Popular coding agents typically rely on either static analysis of the code or iterative test-fix cycles, which is akin to trial and error debugging. We posit that there is a wealth of rich runtime information that developers routinely access while debugging code, which agents are currently deprived of due to design limitations. Despite how prevalent debuggers are in modern IDEs and command-line tools, they have surprisingly not made their way into coding agents. In this work, we introduce Debug2Fix, a novel framework that incorporates interactive debugging as a core component of a software engineering agent via a subagent architecture. We incorporate debuggers for Java and Python into our agent framework and evaluate against GitBug-Java and SWE-Bench-Live and achieve >20% improvement in performance compared to the baseline for certain models. Furthermore, using our framework, we're able to make weaker models like GPT-5 and Claude Haiku 4.5 match or exceed the performances of stronger models like Claude Sonnet 4.5, showing that better tool design is often just as important as switching to a more expensive model. Finally, we conduct systematic ablations demonstrating the importance of both the subagent architecture and debugger integration.

Spandan Garg, Vikram Nitin, Yufan Huang

Modern coding agents increasingly delegate specialized subtasks to subagents, which are smaller, focused agentic loops that handle narrow responsibilities like search, debugging or terminal execution. This architectural pattern keeps the main agent's context window clean by isolating verbose outputs (e.g. build logs, test results, etc.) within the subagent context. Typically when agents employ subagents for such tasks, they use frontier models as these subagents. In this paper, we investigate whether a finetuned small language model (SLM) can achieve comparable performance to frontier models in the task of agentic terminal execution. We present Terminus-4B, which is a post-trained Qwen3-4B model via Supervised Finetuning (SFT) and Reinforcement Learning (RL) using rubric-based LLM-as-judge reward, specifically for this task. In our extensive evaluation spanning various frontier models, training ablations and main agent configurations, we find that Terminus-4B is able to reduce the token usage of the main agent by up to ~30% compared to the No Subagent baseline with no impact to agent performance on benchmarks like SWE-Bench Pro and our internal SWE-Bench C# benchmark, which tends to be heavy in verbose execution tasks. Furthermore, Terminus-4B improves key metrics showing the main agent relying on the outputs of the subagent and doing fewer terminal execution tasks by itself. We see that our model not only closes the gap between the Vanilla Qwen model and frontier models like Claude Sonnet / Opus / GPT-5.3-Codex, but often even exceeds their performance.

Shanchao Liang, Spandan Garg, Roshanak Zilouchian Moghaddam

As large language models (LLMs) become increasingly capable and widely adopted, benchmarks play a central role in assessing their practical utility. For example, SWE-Bench Verified has emerged as a critical benchmark for evaluating LLMs' software engineering abilities, particularly their aptitude for resolving real-world GitHub issues. Recent LLMs show impressive performance on SWE-Bench, leading to optimism about their capacity for complex coding tasks. However, current evaluation protocols may overstate these models' true capabilities. It is crucial to distinguish LLMs' generalizable problem-solving ability and other learned artifacts. In this work, we introduce two diagnostic tasks: file path identification from issue descriptions alone and ground truth function reproduction with only the current file context and issue description to probe models' underlying knowledge. We present empirical evidence that performance gains on SWE-Bench-Verified may be partially driven by memorization rather than genuine problem-solving. We show that state-of-the-art models achieve up to 76% accuracy in identifying buggy file paths using only issue descriptions, without access to repository structure. This performance is merely up to 53% on tasks from repositories not included in SWE-Bench, pointing to possible data contamination or memorization. Similar patterns are also observed for the function reproduction task, where the verbatim similarity is much higher on SWE-Bench Verified than on other similar coding benchmarks (up to 35% consecutive 5-gram accuracy on SWE-Bench Verified and Full, but only up to 18% for tasks in other benchmarks). These findings raise concerns about the validity of existing results and underscore the need for more robust, contamination-resistant benchmarks to reliably evaluate LLMs' coding abilities.

Spandan Garg, Benjamin Steenhoek, Yufan Huang

Current benchmarks for evaluating software engineering agents, such as SWE-Bench Verified, are predominantly derived from GitHub issues and fail to accurately reflect how developers interact with chat-based coding assistants in integrated development environments (IDEs). We posit that this mismatch leads to a systematic overestimation of agent's capabilities in real-world scenarios, especially bug fixing. We introduce a novel benchmarking framework that transforms existing formal benchmarks into realistic user queries through systematic analysis of developer interaction patterns with chat-based agents. Our methodology is flexible and can be easily extended to existing benchmarks. In this paper, we apply our testing framework to SWE-Bench Verified, the TypeScript subset of Multi-SWE-Bench and a private benchmark, SWE-Bench C# and transform formal GitHub issue descriptions into realistic user-style queries based on telemetry analysis of a popular chat-based agent interactions. Our findings reveal that existing benchmarks significantly overestimate agent capabilities for some models by >50% over baseline performance for public benchmarks and ~10-16% for our internal benchmark. This work establishes a new paradigm for evaluating interactive chat-based software engineering agents through benchmark mutation techniques.

Spandan Garg, Roshanak Zilouchian Moghaddam, Neel Sundaresan · microsoft-research

Performance bugs are inefficiencies in software that waste computational resources without causing functional failures, making them particularly challenging to detect and fix. While recent advances in Software Engineering agents have shown promise in automated bug fixing, existing benchmarks primarily focus on functional correctness and fail to evaluate agents' abilities to identify and resolve non-functional issues like performance bugs. We introduce PerfBench, a benchmark comprising 81 real-world performance bug-fixing tasks from popular .NET repositories on GitHub. Unlike existing benchmarks that rely on pre-existing test suites, PerfBench features a novel evaluation harness that allows agents to generate their own performance benchmarks and validates fixes by comparing execution metrics collected for developer fix and agent fix. Each task in PerfBench is derived from actual developer fixes linked to performance-related issues, which are then verified by human experts, ensuring real-world relevance. Our evaluation reveals that current state-of-the-art coding agents struggle with performance optimization tasks, with baseline OpenHands agent achieving only a ~3% success rate on our benchmark. We develop OpenHands-Perf-Agent, which incorporates performance-aware tooling and instructions and achieves a ~20% success rate on the benchmark. We show that by ensuring the agent has proper instructions to benchmark its changes and tooling for benchmark output processing, we can improve the agent performance significantly, but room for improvement still remains. PerfBench provides a challenging test set for furthering the capabilities of agents in fixing performance issues.