Benchmark error during building baseline of EVA-02 model
I got the benchmark error during building baseline of EVA-02. EVA-02 model is timm model for image classification. So I used your code uploaded in OwLite's github to reduce the weight of EVA-02 model. But I got the error as shown in the picture below during building baseline of model. This error did not appear in the terminal window and was only visible on the OwLite project page.
In other words, in terminal window, error message didn't appear and the step of 'Polling for benchmark result' did not end and loaded infinitely. Although the onnx file of EVA-02 baseline was created, benchmark failed. I guess the error was caused by importing SplitToSequence node, but I don't know exactly how to solve. Please give me a hint or a solution.
Enviorment:
python: 3.10.12
pytorch: 2.4.0
owlite: 2.2.0
timm: 1.0.11
Code:
I built main.py in github of OwLite example. I linked this code.
-
Hello Melody. Thank you for using OwLite.
Thank you for providing detailed information about the environment in which the issue occurred.
I have just notified our development team about your request, and they will get back to you soon with a response.
Thank you!
0 -
Dear Melody,
Thank you for choosing OwLite and for bringing this issue to our attention.
I checked and found that ONNX parser of TensorRT doesn’t support SplitToSequence operator yet. So you need to fix the code of EVA model. Specifically, you need to change the code to a node that does the same thing as the SplitToSequence node, but is supported by tensorrt. You need to understand how the EVA model works by looking at the inputs of the SplitToSequence node in the already created onnx model. You can use an application such as netron to see the structure of an onnx model.
We appreciate your patience and encourage you to reach out if you need further assistance.0
Please sign in to leave a comment.
Comments
2 comments