[Python] Debugging vLLM with debugpy on WSL fails: AttributeError: module 'ray' has no...

I’m trying to debug my Python code with debugpy in VS Code on WSL (Ubuntu).

Environment:

• 
  Python: 3.12.3 (venv)
  
  
• 
  vLLM: 0.10.1.1+cpu
  
  
• 
  debugpy: ms-python.debugpy-2025.10.0 (from VS Code extension)
  
  
• 
  Model: Tiny LLM downloaded from Hugging Face
  
  
• 
  Running inside WSL2 (Ubuntu)
  

Code (run-model.py):


from vllm import LLM, SamplingParams

def main():
llm = LLM(model="./Tiny-LLM")
params = SamplingParams(temperature=0.7, max_tokens=200)
prompt = "Once upon a time in a magical forest,"

for output in llm.generate([prompt], sampling_params=params):
print(output.outputs[0].text)

if __name__ == "__main__":
main()

When I run this script normally in the terminal, it works and generates text.

But when I run it with debugpy in VS Code, I get:

AttributeError: module 'ray' has no attribute 'is_initialized'


Traceback (excerpt):

File ".../vllm/engine/arg_utils.py", line 1127, in create_engine_config
if is_ray_initialized():
File ".../vllm/ray/lazy_utils.py", line 9, in is_ray_initialized
return ray.is_initialized()
AttributeError: module 'ray' has no attribute 'is_initialized'


launch.json:


{
"version": "0.2.0",
"configurations": [
{
"name": "Debug run-model.py with vllm",
"type": "debugpy",
"request": "launch",
"program": "${workspaceFolder}/run-model.py",
"console": "integratedTerminal",
"cwd": "${workspaceFolder}",
"justMyCode": false,
"env": {
"VLLM_TARGET_DEVICE": "cpu",
"PYTHONPATH": "${workspaceFolder}/../vllm"
}
}
]
}

What I’ve tried so far:

• Verified that the model loads and runs outside of debugpy (works fine).
• Installed and reinstalled different Ray versions (2.9.3, 2.49.1, etc.) — same error.
• Checked that ray.is_initialized() exists when I import ray directly in Python (works outside vLLM).
• Tried running debugpy both in integratedTerminal and externalTerminal.
• Explicitly set environment variables (VLLM_TARGET_DEVICE=cpu, PYTHONPATH=...) in launch.json.

Questions:

1. Why does vLLM call ray.is_initialized() only when running under debugpy?
2. Is this a Ray version mismatch, or something about vLLM’s CPU mode with debugpy?
3. What’s the correct way to set up debugpy with vLLM on CPU in WSL?

Continue reading...