Error with uploaded JSONL dataset: The specified file reference must point to a completed file import.

Question

I found an error with creating a customized model process: The specified file reference must point to a completed file import.

Noted:

• Both validation.jsonl and training.jsonl can be downloaded as a zip file here.
• both JSONL file, validation.jsonl, and training.jsonl, uploaded via Azure Data Studio. But I discovered that using it via Azure Blob Storage gave the same error.
• This error also happened with ada and babbage model

Step to reproduce:

1. Open Azure OpenAI Studio > Management > Models
2. Click Create Customized Model
   
3. Select the base model type and provide the model's suffix
   
4. In the training data section, click Local file tab
   
5. 6. 7. Click browse a file, and I choose the training.jsonl file in my MacOS, then click Upload file. The windows showed the selected training file as green, and I clicked the Next button to proceed to the next step.
         
6. 9. 10. 11. I did the same process for validation.jsonl in the validation data section

12. 13. I leave the advanced options as the default, Click next to proceed

14. 15. The summary section showed, then I clicked save a close.

16. In 3 minutes, I got the error from Azure OpenAI Studio within 5 seconds

and if I tried to upload both JSONL files from File management menu, it will show Failed for each:

Answer 1

teerasej Have you tried to use the OpenAI CLI data preperation tool to check if there is any reformatting required before uploading the file to the studio and trying again?

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Answer 2

I can no longer find a CLI data prep tool, but in my case, the problem was that the JSONL file did not have each instance on a row, rather the JSONs were formatted and indented. Once I removed the indenting, it was accepted by the tool