Content hash
A short fingerprint computed from a file's or dataset's contents: identical contents give the same hash, and any change gives a different one.
Six weeks from now someone asks which examples trained the model that's live. Git tells you which code ran, but the training file was regenerated twice since then. The fix is cheap: give every dataset a content hash, a short fingerprint computed from its contents, and record it with every training run and eval result. Same rows, same hash; change one character in one row and the hash changes completely.
Try it now
The same checklist works on a tiny GPT and on a real open model.
Learn it properly
Fine-tune an open model · Run a real fine-tune and evaluate the result. About 45 minutes.
Other terms from the same lesson
- rejection-sampling fine-tuning: Generating many answers, keeping only the ones that pass a check, and fine-tuning on those.
Never trained an AI?
Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account.