Before Submitting a New Issue

Oct 13, 2025 by miodeya.com 76 views

Your Current Environment

As per the comment, the environment is set up with the following configuration:

VLLM_USE_V1=1: This environment variable is set to 1, indicating that the V1 version of the VLLM (Vision and Language Model) is being used.
pytest -v -s models/decoder_only/vision_language/test_models.py::test_video_models -k qwen2_vl: This command is used to run the test cases for the video models in the test_video_models module.

Describe the Bug

When running the test cases for the video models using the pytest command, the following error is encountered:

ERROR 03-10 01:17:44 [core.py:324] EngineCore hit an exception: Traceback (most recent call last):
ERROR 03-10 01:17:44 [core.py:324]   File "/home/rshaw/vllm/vllm/v1/engine/core.py", line 317, in run_engine_core
ERROR 03-10 01:17:44 [core.py:324]     engine_core.run_busy_loop()
ERROR 03-10 01:17:44 [core.py:324]   File "/home/rshaw/vllm/vllm/v1/engine/core.py", line 351, in run_busy_loop
ERROR 03-10 01:17:44 [core.py:324]     outputs = step_fn()
ERROR 03-10 01:17:44 [core.py:324]               ^^^^^^^^^
ERROR 03-10 01:17:44 [core.py:324]   File "/home/rshaw/vllm/vllm/v1/engine/core.py", line 174, in step
ERROR 03-10 01:17:44 [core.py:324]     output = self.model_executor.execute_model(scheduler_output)
ERROR 03-10 01:17:44 [core.py:324]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-10 01:17:44 [core.py:324]   File "/home/rshaw/vllm/venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
ERROR 03-10 01:17:44 [core.py:324]     return func(*args, **kwargs)
ERROR 03-10 01:17:44 [core.py:324]            ^^^^^^^^^^^^^^^^^^^^^
ERROR 03-10 01:17:44 [core.py:324]   File "/home/rshaw/vllm/vllm/v1/worker/gpu_model_runner.py", line 922, in execute_model
ERROR 03-10 01:17:44 [core.py:324]     self._update_states(scheduler_output)
ERROR 03-10 01:17:44 [core.py:324]   File "/home/rshaw/vllm/vllm/v1/worker/gpu_model_runner.py", line 360, in _update_states
ERROR 03-10 01:17:44 [core.py:324]     MRotaryEmbedding.get_input_positions_tensor(
ERROR 03-10 01:17:44 [core.py:324]   File "/home/rshaw/vllm/vllm/model_executor/layers/rotary_embedding.py", line 1010, in get_input_positions_tensor
ERROR 03-10 01:17:44 [core.py:324]     video_second_per_grid_t = second_per_grid_ts[video_index]
ERROR 03-10 01:17:44 [core.py:324]                               ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
ERROR 03-10 01:17:44 [core.py:324] IndexError: list index out of range
ERROR 03-10 01:17:44 [core.py:324] 
CRITICAL 03-10 01:17:44 [core_client.py:260] Got fatal signal from worker processes, shutting down. See stack trace above for root cause issue.
Killed

Before Submitting a New Issue

Before submitting a new issue, make sure to:

Search for relevant issues: Check if the issue has already been reported and fixed in the documentation or on the issue tracker.
Ask the chatbot: The chatbot living at the bottom right corner of the documentation page can answer lots of frequently asked questions.

Possible Causes of the Bug

Based on the stack trace, the possible causes of the bug are:

IndexError: The IndexError exception is raised when trying to access an index that is out of range. In this case, the video_index variable is trying to access an index in the second_per_grid_ts list that does not exist.
Model Executor: The model executor is trying to execute the model, but it is encountering an error. This could be due to a problem with the model itself or with the way it is being executed.

Steps to Reproduce the Bug

To reproduce the bug, follow these steps:

Set up the environment: Set up the environment with the V1 version of the VLLM and the pytest command.
Run the test cases: Run the test cases for the video models using the pytest command.
Observe the error: Observe the error message and the stack trace to determine the cause of the bug.

Expected Behavior

The expected behavior is that the test cases for the video models should run successfully without encountering any errors.

Actual Behavior

The actual behavior is that the test cases for the video models encounter an IndexError exception when trying to access an index in the second_per_grid_ts list that does not exist.

Workaround

A possible workaround for this bug is to modify the video_index variable to access a valid index in the second_per_grid_ts list. However, this may not be a feasible solution, as it may require significant changes to the code.

Conclusion

Q: What is the issue with the qwen-vl model?

A: The qwen-vl model is experiencing an issue with video inputs, resulting in an IndexError exception when trying to access an index in the second_per_grid_ts list that does not exist.

Q: What is the expected behavior of the qwen-vl model?

A: The expected behavior of the qwen-vl model is that it should run successfully without encountering any errors when processing video inputs.

Q: What is the actual behavior of the qwen-vl model?

A: The actual behavior of the qwen-vl model is that it encounters an IndexError exception when trying to access an index in the second_per_grid_ts list that does not exist.

Q: What are the possible causes of the bug?

A: The possible causes of the bug are:

IndexError: The IndexError exception is raised when trying to access an index that is out of range. In this case, the video_index variable is trying to access an index in the second_per_grid_ts list that does not exist.
Model Executor: The model executor is trying to execute the model, but it is encountering an error. This could be due to a problem with the model itself or with the way it is being executed.

Q: How can I reproduce the bug?

A: To reproduce the bug, follow these steps:

Set up the environment: Set up the environment with the V1 version of the VLLM and the pytest command.
Run the test cases: Run the test cases for the video models using the pytest command.
Observe the error: Observe the error message and the stack trace to determine the cause of the bug.

Q: What is the workaround for this bug?

A: A possible workaround for this bug is to modify the video_index variable to access a valid index in the second_per_grid_ts list. However, this may not be a feasible solution, as it may require significant changes to the code.

Q: How can I fix the bug?

A: To fix the bug, you will need to identify the root cause of the issue and modify the code accordingly. This may involve modifying the video_index variable to access a valid index in the second_per_grid_ts list, or modifying the model executor to handle the error correctly.

Q: Where can I find more information about the bug?

A: You can find more information about the bug by checking the documentation and the issue tracker for the VLLM project.

Q: How can I get help with fixing the bug?

A: You can get help with fixing the bug by reaching out to the VLLM community or by seeking assistance from a developer who is familiar with the VLLM project.

Q: What are the next steps to take?

A: The next steps to take are to:

Identify the root cause of the issue: Determine the cause of the bug and identify the necessary changes to fix it.
Modify the code: Modify the code to fix the bug and ensure that it is working correctly.
Test the code: Test the code to ensure that it is working correctly and that the bug has been fixed.

Conclusion

In conclusion, the bug is caused by an IndexError exception when trying to access an index in the second_per_grid_ts list that does not exist. The expected behavior is that the test cases for the video models should run successfully without encountering any errors. A possible workaround for this bug is to modify the video_index variable to access a valid index in the second_per_grid_ts list. However, this may not be a feasible solution, as it may require significant changes to the code. To fix the bug, you will need to identify the root cause of the issue and modify the code accordingly.