Jack Pan

#fine-tuning

2 posts

Phase 2: what counts as ground truth

When you harvest Label Studio exports to fine-tune the next model, treating *everything* in the export as ground truth is how you train the model on its own predictions. The filters worth applying before a single byte goes into a training set.

computer visionlabel studiofine-tuning

Phase 2: don't retrain on every export

After every batch of human-corrected episodes gets exported from Label Studio, the temptation is to retrain immediately. The reasons not to, and a cheap cadence trigger that actually fires when retraining will help.

fine-tuninghitlml ops