Fine-tuning dataset validator
Checks AI training files for mistakes before you pay to train on them.
- Status
- Local utility
- Year
- 2026
- Role
- Solo, AI-assisted
- Area
- Developer tools
In plain English
Fine-tuning a model means uploading a file of example conversations and paying for a training run. One malformed example can waste that money. This small tool catches those problems first.
How it works
- Read the training file as JSONL (one example conversation per line).
- Check that each conversation's speakers appear in a valid order, such as user then assistant.
- Confirm each example has assistant turns the model can actually learn from.
- Validate tool-call arguments and warn about examples that may be too long.
What I did
I put this utility together solo, with AI assistance.
What I learned
Catching a broken training file before you pay for a training run is the cheapest win in machine learning.