How can I limit the RAM that is used?

[lesshaste]

In my optimization the llm eventually chooses code that uses all the RAM which kills my computer. What is the best way to limit the RAM so that the evolved code is killed and the openevolve run carries on?

[codelion]

Run evolved code in a subprocess with memory limits.

# In your custom evaluator.py
import subprocess
import resource
import tempfile
import os

def evaluate(program_path: str) -> dict:
    """Evaluate with 2GB memory limit"""
    memory_limit_mb = 2048  # Adjust as needed

    wrapper = f'''
import resource
# Set memory limit
resource.setrlimit(resource.RLIMIT_AS, ({memory_limit_mb * 1024 * 1024}, {memory_limit_mb * 1024 * 1024}))

# Run evaluation
exec(open("{program_path}").read())
# ... your evaluation logic ...
'''

    try:
        with tempfile.NamedTemporaryFile(mode='w', suffix='.py', delete=False) as f:
            f.write(wrapper)
            wrapper_path = f.name

        result = subprocess.run(
            ['python', wrapper_path],
            capture_output=True,
            timeout=300
        )
        os.unlink(wrapper_path)

        if result.returncode != 0:
            return {"error": 0.0, "memory_exceeded": True}

        # Parse and return your metrics
        return parse_metrics(result.stdout)

    except subprocess.TimeoutExpired:
        return {"error": 0.0, "timeout": True}

[lesshaste]

Thank you.