When following the Docs for "Auto label your data" with OpenAI GPT, one always run into an error: "Auto-labeling job failed" Unfortunately there is no more information for further debugging, it's all you get. I tried may different settings, including different Regions (all of them are listed as available for that feature). So either the feature is broken, or the documentation is missing out an essential point. Steps to recreate: Create a language resource Add a storage container with some sample txt Create a Azure OpenAI Resource Go to language studio and create a new Project using all those resources Add entities to the project Run autolabeling with GPT Run ends up with and Erorro: "Auto-labeling job failed" without any further information. Would be really need if it least one could know what is the issue here. Does someone has an idea?

Azure Autolabeling of custom NER in Languade Studio not working

Aditunoe 0

When following the Docs for "Auto label your data" with OpenAI GPT, one always run into an error:

"Auto-labeling job failed"

Unfortunately there is no more information for further debugging, it's all you get. I tried may different settings, including different Regions (all of them are listed as available for that feature). So either the feature is broken, or the documentation is missing out an essential point.

Steps to recreate:

Create a language resource
Add a storage container with some sample txt
Create a Azure OpenAI Resource
Go to language studio and create a new Project using all those resources
Add entities to the project
Run autolabeling with GPT
Run ends up with and Erorro: "Auto-labeling job failed" without any further information.

Would be really need if it least one could know what is the issue here.

Does someone has an idea?

romungi-MSFT 42,311 Reputation points Microsoft Employee

2024-01-23T10:55:57.85+00:00

@Aditunoe Could you try to add diagnostic logging to your language & OpenAI resource? This is a setting from azure portal on the Monitoring blade of the resource. This can help get logs to debug the issue. Also, since you are using storage and OpenAI resource deployment are you having any network limitations on the storage account and are there any limits on the tokens that are available for your deployment?

Aditunoe 0

Hey,

thanks for your fast answer. I followed your advice and turned on the diagnostic logs by creating a Log Analytics workspace and adding diagnostic logs settings to each, language & OpenAI resource. Here are the recent logs after a failed job:

key	value
TimeGenerated [UTC]	2024-01-28T20:34:25.668Z
ResourceProvider	MICROSOFT.COGNITIVESERVICES
ResourceType	ACCOUNTS
OperationName	AnalyzeText_Internal_GPT_GetGptJobStatus
DurationMs	69
ResultSignature	200

key	value
OperationName	GetBlob
AuthenticationType	OAuth
StatusCode	200
StatusText	Success

To your other question, I created storage and OpenAI Ressource without any network limitations and for the token, i used gpt-turbo but the token limit cant be an issue, as I tried it with a one liner text.

enter image description here

Can I provide anything else specific that may help?

romungi-MSFT 42,311 Reputation points Microsoft Employee

2024-02-02T07:32:44.3466667+00:00

@Aditunoe Unfortunately, i cannot lookup this activity ID to check why it failed. The diagnostic logs indicate a success so I think looking up the job activity ID is required. Do you have a support plan for your subscription? Thanks!!
Aditunoe 0 Reputation points

2024-02-03T09:44:49.92+00:00

@romungi-MSFT Yes, we actually do have a Standard support plan. Should we try to proceed there?
romungi-MSFT 42,311 Reputation points Microsoft Employee

2024-02-05T05:22:00.5966667+00:00

Yes @Aditunoe . That should help.