Gpt 4o finetuning stuck at running for more than 24 hours

Hi Pushkar Gaikwad,

I understand that your fine-tuning job for the GPT-4o model has been stuck in the "Training" state for over 24 hours. There are a few potential reasons this might happen and some steps you can take to troubleshoot it.

First, even though your dataset is relatively small (around 1.4 MB), the fine-tuning job could be stuck due to high demand or limited compute resources in your region East US 2. Jobs are often queued behind others, especially during peak usage times. It’s a good idea to refresh the Azure AI Studio portal and check for any updates to the job status.

Although the file size is small, it's still important to make sure your dataset meets the required formatting and quality standards for fine-tuning. A poorly structured or invalid dataset might result in the job stalling without an obvious error.

If the job continues to show no progress, I’d recommend canceling and resubmitting it. you might try validating the data again and adjusting the configuration slightly when resubmitting—this can sometimes help bypass unexpected edge issues.

Azure’s compute resources can occasionally be under high load, particularly in popular regions. You can check the Azure Status Page to see if there are any service incidents or regional capacity issues impacting East US 2.

Also you can refer Check the status of your custom model,

Troubleshooting for Azure OpenAI fine-tuning.

I Hope this helps. Do let me know if you have any further queries.

Thank you!

Hi @Pavankumar Purilla ,

I have the same problem and not first time I tried a fine tuning before with similar size and it stuck for over 24h
and now I am trying with a small dataset about 512k training file
and seems to be stuck as well

I used the UI for training first time then rest api for second time same results

I verified that dataset format as it produces results on OpenAi platform and fine tuning is working normally there

can you check if it is not a problem from your side?

Mine was completed after 36 hours, so I suggest you wait a bit.

Hi Omar Elnahas,
Thank you for reaching out. Fine-tuning jobs can sometimes take longer depending on system load and region capacity, even with small datasets. Since your job is still in the “Running” state and hasn’t failed, it’s likely still processing or queued. Could you please wait a little longer to allow it to complete.