Empty Model is creating while using Train Custom Model Asyc Post Request

Sahil Khurana 1

We are trying out Creating Custom Model Asyc service and using Blob SAS URI as source and putting other required paramenters like
"sourceFilter": {
"prefix": "",
"includeSubFolders": false
},
"useLabelFile": false

When we keep useLabelFile:False
The post request cerates an empty model

when we change useLabel:True , Model is not created howver resopnse came as 201.

1 answer

romungi-MSFT 43,696 Reputation points Microsoft Employee

2022-11-03T11:31:04.46+00:00
@Sahil Khurana I think you are using the API version v2.1 with the source string, which contains the location of your training files, and these training files might not be accessible.
Are you using the format as mentioned in your request body?

{ "source": "string", "sourceFilter": { "prefix": "string", "includeSubFolders": false }, "useLabelFile": false }

Is this source an azure storage account with the required permissions?
I think you can configure the same source string with the FOTT labeling tool and check if the training documents are loading before you submit the training request through the tool.
Please sign in to rate this answer.
Sahil Khurana 1 Reputation point

2022-11-03T13:27:46.057+00:00

Yes this is azure storage account with required permission. With same Blob Storage SAS Uri we are able to create models using Form Recognizer.

Sahil Khurana 1 Reputation point

2022-11-03T18:39:04.247+00:00

We used the same Blob Container SAS URI in fott-2 Tool, and we were able to load images and create label and models.
However, when we are trying to create model using "Async" service, always an empty model gets created with random id and no fields.

romungi-MSFT 43,696 Reputation points Microsoft Employee

2022-11-04T15:01:27.537+00:00

@Sahil Khurana If the same SAS URL works with FOTT tool then the same should work with the REST API. If you are not really using any subfolders or a particular label file format to train the model then you could use only the source field in your request body and check if it works. Adding additional fields might be causing your request to find files that are not available and creating a dummy model. A request body like the following might work with the same SAS. Please ensure that your SAS token is not expired though.

curl -i -X POST "https://{endpoint}/formrecognizer/v2.1/custom/models" -H "Content-Type: application/json" -H "Ocp-Apim-Subscription-Key: {key}" --data-ascii "{ 'source': '{SAS URL}'}"

Body:

{ 'source': '{SAS URL}' }

Sahil Khurana 1 Reputation point

2022-11-14T09:09:18.083+00:00

i tried both ways thatyou mentioned above.
With Curl command i cerated post request and below was the response along with dummy/empty model

Sahil Khurana 1 Reputation point

2022-11-14T09:10:12.597+00:00

HTTP/1.1 201 Created
Sign in to comment

Share via

Empty Model is creating while using Train Custom Model Asyc Post Request

1 answer