Errors During Fine-Tuning and Inference in Surya OCR: Missing Input Images & Undefined SPECIAL_TOKENS

[Atharv-Bhujade]

I need some assistance with an issue I'm facing. While fine-tuning Surya OCR, I encounter two recurring errors:

During the fine-tuning process, I get the following error:
AssertionError: Image tokens were present in the input but no input images were provided.

If the training does complete successfully, I encounter this error during inference of the checkpoint:
NameError: name 'SPECIAL_TOKENS' is not defined.

Any guidance would be greatly appreciated!

[tarun-menta]

Hi @Atharv-Bhujade. Apologies for the inconvenience. We updated the model and this script has gone out-of-sync. Will fix shortly