How can I increase the filesize-limit of 500MB for training data for fine-tuning in Azure OpenAI?

Question

When trying to fine-tune a base model via Azure OpenAI Studio I stumbled across the filesize-limit of 200MB for direct upload and 500MB for uploading through a blob, respectively.

Splitting my training data into chunks meeting these limits also does not help, since I can only pick a single data file of all my uploaded data files. This is very unfortunate. How can I increase this limit? Would it help to create the fine-tuning-job differently, e.g. Python SDK?

Answer 1

Hauke Kraemer Greetings & Welcome to Microsoft Q&A forum!

Your understanding is correct regarding the limits.

Training data files must be formatted as JSONL files, encoded in UTF-8 with a byte-order mark (BOM). The file must be less than 512 MB in size.

Splitting my training data into chunks meeting these limits also does not help, since I can only pick a single data file of all my uploaded data files. This is very unfortunate. How can I increase this limit? Would it help to create the fine-tuning-job differently, e.g. Python SDK?

One thing you could try is to submit a service request for increasing the limits.

Do let me know if you have any further queries.