atropos/environments/kernelbench_env

KernelBench Environment Setup Instructions

Prerequisites

Before running kernelbench_env.py, you need to install KernelBench:

1. Install KernelBench from source:

   pip install git@github.com:ScalingIntelligence/KernelBench.git
   cd KernelBench
   pip install -r requirements.txt
   pip install -e .
   cd -
   

2. Set variables at the top of the script:

   • KERNELBENCH_LEVEL: The difficulty level (1-3)
   • KERNELBENCH_PROBLEM_NUMBER: The specific problem number to solve
   • KERNELBENCH_DIR: The absolute path to your KernelBench install

3. Run kernelbench_env.py as you would any other environment:

   python kernelbench_env.py serve --slurm False
   
   

These environment variables will be used to configure the evaluation environment.

Citations

Baronio, C., Marsella, P., Pan, B., & Alberti, S. (2025 May 6). Kevin‑32B: Multi‑Turn RL for Writing CUDA Kernels. Cognition AI Blog. Retrieved May 16, 2025, from https://cognition.ai/blog/kevin-32b