Valid JSONL can still make broken fine-tuning data
Parallelogram is a tool for checking datasets. A data file can be valid JSONL but still be bad for training. Common problems include the wrong role order, empty assistant answers, repeated examples, overflow, and strange text encoding issues.
A public check found that the website had HSTS but was missing several basic . The missing pieces included CSP, frame protection, nosniff, Referrer-Policy, robots.txt, and security.txt. Those have now been added, along with -Policy, a sitemap, and SECURITY.md in the .
The browser demo still makes no while checking a dataset.
Key points
- Valid JSONL does not mean the data is safe or useful for .
- The tool checks issues like bad role order, empty answers, duplicates, overflow, and encoding problems.
- The browser demo checks datasets without making .
- The website now includes CSP, Referrer-Policy, security.txt, and other basic security files.
- For AI agent work, catching bad early can reduce wasted testing and training cost.