Open haseebrj17 opened 1 month ago
@czczup @opengvlab-admin Can you please take a look at the above error!
Hello, this seems to be the wrong shape of the image tensor you entered, please check it
@czczup What shape image tensor does the input require, and can you please elaborate on your previous comment, I would appreciate it!
Checklist
Describe the bug
When processing a PDF file, the script extracts images from the PDF and passes them to the model to generate captions and descriptions. However, the pixel_values tensor is incorrectly formed, resulting in a shape of [1, 1]. This causes a ValueError in the vision model’s forward pass, indicating that the pixel_values size is incorrect.
Reproduction
Use the following script for DataPipelineLLM.py:
Environment
Error traceback