How to train Document Understanding Model in AI-Fabric

How to train Document Understanding Model in AI-Fabric?

Below are steps that needs to be followed/taken-care for training the Document Understanding Out Of Box package in AI-Fabric:

Note: Document Understanding Package must be trained before deployment to ensure that ML-Skill creation will  not fail.

1. Create a ML package using the OOB Document understanding model in AI-Fabric.Refer this link.

2. Then dataset needs to be created in the AI-Fabric  under the  “Data sets” tab. Refer this link .

Below are the steps to be followed/taken-care while preparing dataset for DU model:
a) Data labelling needs to be performed on all the documents using  "Data Manager". Steps given in this link needs to be followed for installing Data Manager:

b) At least 25 documents needs to be labelled in order to train the model and each regular field should be labelled on at at least 10 documents. Refer this link for using Data Manager:

c) Below export requirements should be met. Refer this link

d) Now while uploading dataset, files should be uploaded in exact format mentioned in the link :
3. Once the data set is uploaded, pipeline needs to created and correct dataset needs to be tagged in pipeline.Refer below link to know more about pipeline creation  here .

4. After successful training , new package version will get generated and this needs to be used for creating the ML-Skill. Refer this link.
1 Like