AB
AB
AiBoss

All Categories

project

FrogBoss - Microsoft's open-source programming model suite

FrogBoss is a series of programming models developed by Microsoft Research. The models are trained using an innovative and sophisticated vulnerability generation method (BugPilot), which involves having a Language Model Agent (LLM) add new functionality to the codebase...

What is FrogBoss?

FrogBoss is a series of programming models developed by Microsoft Research. The models are trained using an innovative, complex vulnerability generation method (BugPilot), which inadvertently introduces vulnerabilities by having a Language Model Agent (LLM) add new features to the codebase, generating more challenging and diverse training data. The FrogBoss 32-parameter model achieved a pass@1 accuracy of 54.6% in the SWE-Bench-Verified benchmark, becoming the latest benchmark in the field. The model's superior performance demonstrates higher data efficiency and better adaptability to real-world software development scenarios.

FrogBoss's main functions

  • High-efficiency vulnerability repairThe model can quickly identify and fix complex vulnerabilities in code, improving software quality and stability.
  • Code optimization and improvementBy analyzing existing code, we can propose optimization suggestions to improve code efficiency and maintainability.
  • Automated testing supportThe model can generate test cases to verify whether code changes introduce new problems and ensure that the software functions normally.
  • Improved development efficiencyBy automating tedious tasks, repetitive work by developers can be reduced, accelerating the software development cycle.

FrogBoss's technical principles

  • Complex vulnerability generation (BugPilot):
    • FeatAdd methodThis method inadvertently introduces vulnerabilities by having a Language Model Agent (LLM) add new features to the codebase. It simulates vulnerabilities arising from feature extensions in real-world software development, generating vulnerabilities that are more diverse and complex.
    • BugInstruct methodIn contrast, having the proxy directly introduce vulnerabilities generates simpler vulnerabilities that differ significantly from real-world scenarios.
  • High-quality data trainingHigh-quality vulnerability data generated using the FeatAdd method, combined with other datasets (such as R2E-Gym and SWE-Smith), is trained using supervised fine-tuning (SFT) and reinforcement learning (RL). This training approach enables FrogBoss to excel in solving complex software engineering tasks while improving data utilization efficiency.

FrogBoss project address

  • Project official website: https://microsoft.github.io/debug-gym/blog/2025/10/bug-pilot/
  • HuggingFace model library:
    • https://huggingface.co/microsoft/FrogBoss-32B-2510
    • https://huggingface.co/microsoft/FrogMini-14B-2510
  • arXiv technical paperhttps://arxiv.org/pdf/2510.19898

FrogBoss Application Scenarios

  • Code debugging and vulnerability patchingFrogBoss can automatically detect code vulnerabilities and generate fix suggestions, significantly improving software stability and development efficiency.
  • Improve software development efficiencyFrogBoss helps developers quickly achieve high-quality code by providing accessibility features and optimizing code.
  • Automated testingFrogBoss can automatically generate test cases and analyze test results, facilitating more comprehensive software testing.
  • Code review and quality assuranceQuickly identify potential problems during code reviews to ensure software meets high-quality standards.
  • Education and TrainingAs an educational tool, FrogBoss provides code examples and error analysis to help students and developers improve their programming skills.
Back