kernrl - GPU Kernel Optimization Environment

RL environment for training LLMs to write optimized GPU kernels.

API Endpoints

• POST /reset - Start a new episode
• POST /step - Submit kernel code
• GET /state - Get current state
• GET /health - Health check
• GET /problems - List available problems

Note

This is a demo instance running on CPU. Full kernel evaluation requires GPU.

For GPU evaluation, run locally with Docker:

docker run --gpus all -p 8000:8000 kernrl

Links

• API Documentation (Swagger)
• OpenEnv PR
• Training Materials