OCR approach #1
-
|
Hey, thank you for the great work. Would you be able to tell more about recognizing text in document - what tools, or software did you use? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
We used a proprietary OCR model custom-compiled with TensorRT for the Blackwell architecture. It took approximately two weeks of trial and error to identify the optimal settings, after which we achieved massive parallelism that enabled us to process all documents within 72 hours using dozens of B300 chips. Unfortunately, I’m not able to share further details about the model architecture due to IP constraints - it is an ongoing research project. |
Beta Was this translation helpful? Give feedback.
We used a proprietary OCR model custom-compiled with TensorRT for the Blackwell architecture. It took approximately two weeks of trial and error to identify the optimal settings, after which we achieved massive parallelism that enabled us to process all documents within 72 hours using dozens of B300 chips. Unfortunately, I’m not able to share further details about the model architecture due to IP constraints - it is an ongoing research project.