No Sample Submissions? Here’s How to Fine-Tune

Fine-tuning is an incredibly valuable tool for instructors on TimelyGrader. By uploading between one and three sample submissions before grading, you can easily adjust the assignment setup to ensure grading and feedback suggestions align with your standards and expectations. The result is faster grading, more meaningful feedback, and a process that better reflects how you teach and evaluate student work.

But what if you're grading a brand-new assignment and don’t have any sample submissions yet?

Let’s walk through how to approach this, both for the assignment you're grading now and for future assignments.

What to Avoid Doing

We understand that if you’re teaching a new course or introducing a new assignment, you may not have previous student submissions to fine-tune with. In those cases, it might be tempting to use student submissions from the current assignment to fine-tune. However, we strongly advise against this. 

Here’s why: fine-tuning updates the grading rubric, grading instructions, and feedback instructions the AI uses as context for the suggestions it provides. If you’ve already uploaded and graded submissions, duplicating the assignment and grading with a fine-tuned assignment for the remaining submissions can lead to inconsistent assessment, so it’s best to avoid this approach to ensure a fair assessment process.

What You Can Do Instead

If you don’t have sample submissions right now, there are a couple of things you can do:

1. Create your own sample submissions: Consider leveraging your assignment details to generate sample submissions with the help of AI. This method allows you to create a set of representative submissions to use for fine-tuning before grading actual student work. This ensures that your grading setup remains consistent and fair for every student.

2. Perfect the setup for next time: Once you finish grading this assignment, duplicate it and select a few representative student submissions to use as samples. Then, fine-tune the duplicated version. Next time you’re grading this assignment, you’ll have a ready-to-go setup that’s already customized to your grading preferences.

Remember: For best results, use high, medium, and low-scoring sample submissions, this will enable you to preview grading/feedback suggestions across a range of performance. Keep this in mind if you’re creating sample submissions or selecting them from past student submissions.

Hopefully, this gives you some ideas of how to get started if you don’t currently have sample submissions, but if you have any questions, you can reach out to our team at [email protected], and we’d be happy to help!