Open ParasSharmaIISL opened 1 month ago
Thank you for reporting the issue. To unblock the issue, loading the saved model into the inference script might be a good idea. Can you elaborate a bit more on what you mean by not working on inferentia chips? Is there a compiler error or an execution failure?
The calamari model is trained on tensorflow 2.7.0 which utilises tfaip library and the lowest tensorflow_neuron version 2.7.4 uses a higher keras version. Due to this version mismatch I am unable to load my (calamari-ocr) keras model to the python environment with tensorflow-neuron and not able to trace the model. If I save this model in a tensorflow saved model, the traced model's computation graph only works on CPU and not inferntia chips. This is the repository "https://github.com/Calamari-OCR/calamari".