arXiv subjects

Shubham Gandhi

Publications and source records attributed to Shubham Gandhi.

11 recordsLinked to original sources

arXiv2026

DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training

Reinforcement Learning from Verifiable Rewards works well when a task has a programmatic checker, but most long-horizon agent domains have none. We work in the outcome-blind setting, where ground-truth success signals are not available. Multi-criteria rubrics are a popular way to supply such a reward; they are scored once per trajectory, but a single scalar is a poor signal across tens of steps. We propose DRACO: Distributing Rubric-based Advantage for Credit Optimization. It generates rubrics dynamically during training to track the policy's evolving capability, scores those rubrics once per completed trajectory, and redistributes that judgment over the steps responsible for annotated rubrics to produce differentiated per-step advantages in GRPO. The redistribution is closed-form and does not introduce any trained attribution module. On AppWorld, DRACO gains 15.9 points over the base model and 5.3 points over GRPO trained with a sparse ground-truth reward, despite not using any verifiers itself. On out-of-domain Tau-Bench, it gains 5.3 points over the base model even without a frontier judge, beating both ground-truth-reward training and other rubric-based training settings. The code for DRACO is available at https://github.com/IBM/draco.

cs.AI↗

arXiv2026

Steer, Don't Solve: Training Small Critic Models for Large Code Agents

Coding tasks are typically complicated and require multiple capabilities, ranging from high-level planning to low-level implementation. While coding agents are optimized for the joint capabilities, individual capabilities such as high-level planning may have different optima and remain a major bottleneck. To address this challenge, we train a separate critic model that is specialized in high-level planning to steer the coding agent in inference. We construct SFT and DPO data to train the critic model to identify errors made by the coding agent and provide correct and clear high-level guidance without generating concrete actions. Experiments show that our fine-tuned 4B and 8B critic models significantly improve the performance of 6 larger coding agents (e.g., improving the resolved rates of GLM-4.7-Flash-30B-A3B and GPT-OSS-120B by 16.0% and 14.4% on SWE-Bench Verified). The critic model also reduces the total inference costs for some coding agents by solving tasks in fewer steps (e.g., reducing the per-example inference cost for GPT-OSS-20B from \$0.07 to \$0.03). Code: https://github.com/shubhamrgandhi/critic-training

cs.SE↗

arXiv2026

Hybrid-Gym: Training Coding Agents to Generalize Across Tasks

When assessing the quality of coding agents, predominant benchmarks focus on solving single issues on GitHub, such as SWE-Bench. In contrast, in real use, these agents solve more various and complex tasks that involve other skills such as exploring codebases, testing software, and designing architecture. In this paper, we first characterize some transferable skills that are shared across diverse tasks by decomposing trajectories into fine-grained components, and derive a set of principles for designing auxiliary training tasks to teach language models these skills. Guided by these principles, we propose a training environment, Hybrid-Gym, consisting of a set of scalable synthetic tasks, such as function localization and dependency search. Experiments show that agents trained on our synthetic tasks effectively generalize to diverse real-world tasks that are not present in training, improving a base model by 25.4% absolute gain on SWE-Bench Verified, 7.9% on SWT-Bench Verified, and 5.1% on Commit-0 Lite. Hybrid-Gym also complements datasets built for the downstream tasks (e.g., improving SWE-Play by 4.9% on SWT-Bench Verified). Code available at: https://github.com/yiqingxyq/Hybrid-Gym.

cs.SE↗

arXiv2025

When Agents go Astray: Course-Correcting SWE Agents with PRMs

Large Language Model (LLM) agents are increasingly deployed for complex, multi-step software engineering (SWE) tasks. However, their trajectories often contain costly inefficiencies, such as redundant exploration, looping, and failure to terminate once a solution is reached. Prior work has largely treated these errors in a post-hoc manner, diagnosing failures only after execution. In this paper, we introduce SWE-PRM, an inference-time Process Reward Model (PRM) that intervenes during execution to detect and course-correct trajectory-level errors. Our PRM design leverages a taxonomy of common inefficiencies and delivers lightweight, interpretable feedback without modifying the underlying policy. On SWE-bench Verified, closed-source PRMs improve resolution from 40.0% to 50.6% (+10.6 p.p.), with the largest gains on medium and hard tasks. Among feedback strategies, taxonomy-guided PRMs outperform unguided or explicit action-prescriptive variants, increasing success rate while reducing trajectory length. These benefits come at an acceptable added inference cost of as low as $0.2, making PRMs a practical and scalable mechanism for improving SWE agents' reliability and efficiency.

cs.AI↗

arXiv2025

An Empirical Study on Strong-Weak Model Collaboration for Repo-level Code Generation

We study cost-efficient collaboration between strong and weak language models for repository-level code generation, where the weak model handles simpler tasks at lower cost, and the most challenging tasks are delegated to the strong model. While many works propose architectures for this task, few analyze performance relative to cost. We evaluate a broad spectrum of collaboration strategies: context-based, pipeline-based, and dynamic, on GitHub issue resolution. Our most effective collaborative strategy achieves equivalent performance to the strong model while reducing the cost by 40%. Based on our findings, we offer actionable guidelines for choosing collaboration strategies under varying budget and performance constraints. Our results show that strong-weak collaboration substantially boosts the weak model's performance at a fraction of the cost, pipeline and context-based methods being most efficient. We release the code for our work at https://github.com/shubhamrgandhi/codegen-strong-weak-collab.

cs.AI↗

arXiv2025

ResearchCodeAgent: An LLM Multi-Agent System for Automated Codification of Research Methodologies

In this paper we introduce ResearchCodeAgent, a novel multi-agent system leveraging large language models (LLMs) agents to automate the codification of research methodologies described in machine learning literature. The system bridges the gap between high-level research concepts and their practical implementation, allowing researchers auto-generating code of existing research papers for benchmarking or building on top-of existing methods specified in the literature with availability of partial or complete starter code. ResearchCodeAgent employs a flexible agent architecture with a comprehensive action suite, enabling context-aware interactions with the research environment. The system incorporates a dynamic planning mechanism, utilizing both short and long-term memory to adapt its approach iteratively. We evaluate ResearchCodeAgent on three distinct machine learning tasks with distinct task complexity and representing different parts of the ML pipeline: data augmentation, optimization, and data batching. Our results demonstrate the system's effectiveness and generalizability, with 46.9% of generated code being high-quality and error-free, and 25% showing performance improvements over baseline implementations. Empirical analysis shows an average reduction of 57.9% in coding time compared to manual implementation. We observe higher gains for more complex tasks. ResearchCodeAgent represents a significant step towards automating the research implementation process, potentially accelerating the pace of machine learning research.

cs.SE↗

arXiv2025

BudgetMLAgent: A Cost-Effective LLM Multi-Agent system for Automating Machine Learning Tasks

Large Language Models (LLMs) excel in diverse applications including generation of code snippets, but often struggle with generating code for complex Machine Learning (ML) tasks. Although existing LLM single-agent based systems give varying performance depending on the task complexity, they purely rely on larger and expensive models such as GPT-4. Our investigation reveals that no-cost and low-cost models such as Gemini-Pro, Mixtral and CodeLlama perform far worse than GPT-4 in a single-agent setting. With the motivation of developing a cost-efficient LLM based solution for solving ML tasks, we propose an LLM Multi-Agent based system which leverages combination of experts using profiling, efficient retrieval of past observations, LLM cascades, and ask-the-expert calls. Through empirical analysis on ML engineering tasks in the MLAgentBench benchmark, we demonstrate the effectiveness of our system, using no-cost models, namely Gemini as the base LLM, paired with GPT-4 in cascade and expert to serve occasional ask-the-expert calls for planning. With 94.2\% reduction in the cost (from \$0.931 per run cost averaged over all tasks for GPT-4 single agent system to \$0.054), our system is able to yield better average success rate of 32.95\% as compared to GPT-4 single-agent system yielding 22.72\% success rate averaged over all the tasks of MLAgentBench.

cs.MA↗

arXiv2024

A Comparative Analysis of Distributed Training Strategies for GPT-2

The rapid advancement in Large Language Models has been met with significant challenges in their training processes, primarily due to their considerable computational and memory demands. This research examines parallelization techniques developed to address these challenges, enabling the efficient and scalable training of Large Language Models. A comprehensive analysis of both data and model parallelism strategies, including Fully Sharded Data Parallelism and Distributed Data-Parallel frameworks, is provided to assess methods that facilitate efficient model training. Furthermore, the architectural complexities and training methodologies of the Generative Pre-Trained Transformer-2 model are explored. The application of these strategies is further investigated, which is crucial in managing the substantial computational and memory demands of training sophisticated models. This analysis not only highlights the effectiveness of these parallel training strategies in enhancing training efficiency but also their role in enabling the scalable training of large language models. Drawing on recent research findings, through a comprehensive literature review, this research underscores the critical role of parallelization techniques in addressing the computational challenges of training state-of-the-art Large Language Models, thereby contributing to the advancement of training more sophisticated and capable artificial intelligence systems.

cs.DC↗

arXiv2023

Leveraging a Randomized Key Matrix to Enhance the Security of Symmetric Substitution Ciphers

An innovative strategy to enhance the security of symmetric substitution ciphers is presented, through the implementation of a randomized key matrix suitable for various file formats, including but not limited to binary and text files. Despite their historical relevance, symmetric substitution ciphers have been limited by vulnerabilities to cryptanalytic methods like frequency analysis and known plaintext attacks. The aim of our research is to mitigate these vulnerabilities by employing a polyalphabetic substitution strategy that incorporates a distinct randomized key matrix. This matrix plays a pivotal role in generating a unique random key, comprising characters, encompassing both uppercase and lowercase letters, numeric, and special characters, to derive the corresponding ciphertext. The effectiveness of the proposed methodology in enhancing the security of conventional substitution methods for file encryption and decryption is supported by comprehensive testing and analysis, which encompass computational speed, frequency analysis, keyspace examination, Kasiski test, entropy analysis, and the utilization of a large language model.

cs.CR↗

arXiv2023

YOLOv8-Based Visual Detection of Road Hazards: Potholes, Sewer Covers, and Manholes

Effective detection of road hazards plays a pivotal role in road infrastructure maintenance and ensuring road safety. This research paper provides a comprehensive evaluation of YOLOv8, an object detection model, in the context of detecting road hazards such as potholes, Sewer Covers, and Man Holes. A comparative analysis with previous iterations, YOLOv5 and YOLOv7, is conducted, emphasizing the importance of computational efficiency in various applications. The paper delves into the architecture of YOLOv8 and explores image preprocessing techniques aimed at enhancing detection accuracy across diverse conditions, including variations in lighting, road types, hazard sizes, and types. Furthermore, hyperparameter tuning experiments are performed to optimize model performance through adjustments in learning rates, batch sizes, anchor box sizes, and augmentation strategies. Model evaluation is based on Mean Average Precision (mAP), a widely accepted metric for object detection performance. The research assesses the robustness and generalization capabilities of the models through mAP scores calculated across the diverse test scenarios, underlining the significance of YOLOv8 in road hazard detection and infrastructure maintenance.

cs.CV↗

arXiv2023

StaticFixer: From Static Analysis to Static Repair

Static analysis tools are traditionally used to detect and flag programs that violate properties. We show that static analysis tools can also be used to perturb programs that satisfy a property to construct variants that violate the property. Using this insight we can construct paired data sets of unsafe-safe program pairs, and learn strategies to automatically repair property violations. We present a system called \sysname, which automatically repairs information flow vulnerabilities using this approach. Since information flow properties are non-local (both to check and repair), \sysname also introduces a novel domain specific language (DSL) and strategy learning algorithms for synthesizing non-local repairs. We use \sysname to synthesize strategies for repairing two types of information flow vulnerabilities, unvalidated dynamic calls and cross-site scripting, and show that \sysname successfully repairs several hundred vulnerabilities from open source {\sc JavaScript} repositories, outperforming neural baselines built using {\sc CodeT5} and {\sc Codex}. Our datasets can be downloaded from \url{http://aka.ms/StaticFixer}.

cs.SE↗