Task1:single-source-augmentation
For submission
#284478 and
#284476， I used the same code and got different results.
Submission
#284478： it shows submitted but the detailed interface has a score.
Submission #284476：it shows: Evaluation timed out . Please ensure that your code runs within the specified compute budget. Can you provide me with more log information? After all, the same code submission #284478 has results.
Task2: multi-source-augmentation
Submission
#284453:It shows:Gym server stopped unexpectedly. Please contact the admins.
What is the reason? Can you also provide me with the complete log information?
I checked the logs for 284478 and 284476.
The logs of 284476 indeed shows timed out.
2025-05-13 21:17:45.659
2025-05-13 13:17:45.658 | INFO | __main__:run_with_timeout:161 - Timeout 80 reached for batch_generate_response
The logs of 284478 shows that it got the scores. We can further check this.
For 284453, the error most likely looks like this.
2025-05-13 18:36:27.639
[rank0]: RuntimeError: Failed to apply MllamaProcessor on data={'text': '<|image|>', 'images': [<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=9248x6936 at 0x7F7A45574670>]} with kwargs={}
Thanks for your reply.
[rank0]: RuntimeError: Failed to apply MllamaProcessor on data={'text': '<|image|>', 'images': [<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=9248x6936 at 0x7F7A45574670>]} with kwargs={}
This message doesn’t seem to show what went wrong? What do you think is the problem? Can you provide the complete log information?
[rank0]: RuntimeError: Failed to apply MllamaProcessor on data={'text': '<|image|>', 'images': [<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=9248x6936 at 0x7F7A45574670>]} with kwargs={}
Looking forward to your reply. I really don’t know what this log information indicates? Is it because the resolution of the image obtained through the image search API is too high? I noticed in the previous submission that because there was no network connection, the image was obtained from the server’s cache. Because the image was damaged, the evaluation failed.
Submission #284417:
OSError: image file is truncated (61 bytes not processed)
Therefore, I added a judgment condition and this problem no longer occurs.
Is this a common problem? We need to add corresponding judgment conditions to avoid it.
Sorry, but I really have no idea about why the error is there. Maybe you can try to initialize an equally large image locally and try to process it with the same pipeline, and see how it goes. After all, this seems to be more related to the MLLaMA model/processor, than our challenge.
