Checklist

• 1. I have searched related issues but cannot get the expected help.
• 2. The bug has not been fixed in the latest version.
• 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.
• 4. If the issue you raised is not a bug but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 5. Please use English, otherwise it will be closed.

Describe the bug

I tried to use OpenAI models with the frontend language, and installed sglang by using pip install "sglang[openai]".

This correctly installed sglang-v0.4.4.post3 and its dependencies on my machine. However, when I tried to run the openai_example_chat.py example, I got the following error:

  File "/usr/local/opt/miniconda/envs/sglang/lib/python3.12/site-packages/sglang/__init__.py", line 4, in <module>
    from sglang.api import (
  File "/usr/local/opt/miniconda/envs/sglang/lib/python3.12/site-packages/sglang/api.py", line 7, in <module>
    from sglang.lang.backend.base_backend import BaseBackend
  File "/usr/local/opt/miniconda/envs/sglang/lib/python3.12/site-packages/sglang/lang/backend/base_backend.py", line 5, in <module>
    from sglang.lang.interpreter import StreamExecutor
  File "/usr/local/opt/miniconda/envs/sglang/lib/python3.12/site-packages/sglang/lang/interpreter.py", line 34, in <module>
    from sglang.utils import (
  File "/usr/local/opt/miniconda/envs/sglang/lib/python3.12/site-packages/sglang/utils.py", line 28, in <module>
    from sglang.srt.utils import kill_process_tree
  File "/usr/local/opt/miniconda/envs/sglang/lib/python3.12/site-packages/sglang/srt/utils.py", line 48, in <module>
    import torch
ModuleNotFoundError: No module named 'torch'

I found that this was due to the import statement in sglang/utils.py:

from sglang.srt.utils import kill_process_tree

It wrongly introduced a dependency on sglang.srt.utils and further introduced a dependency on torch.

I checked the usage of kill_process_tree in sglang/utils.py and found that it was only used once in the terminate_process function. Then I further checked the change history of sglang/utils.py and found that this commit moved the above import statement from inside the terminate_process function to the beginning of the file.

This change was committed between v0.4.3.post2 and v0.4.3.post3. v0.4.3.post2 can run openai_example_chat.py correctly, while v0.4.3.post3 reported the same error.

Putting the above import statement back inside the terminate_process function can solve this problem.

Reproduction

Just start from a new virtual environment and run pip install "sglang[openai]".

python3 openai_example_chat.py will raise ModuleNotFoundError: No module named 'torch'.

Environment

This script is not really runnable because it imports torch without checking for its existence. For the same reason as #3874.