7 Models, 2 Deployment Targets, and the Part That Actually Mattered Was Calibration
I fine-tuned 7 LoRA adapters across 4 model families for form abandonment classification. Every baseline scored 0.06. The best fine-tune hit 0.96. The interesting part was everything after.
#ai#fine-tuning#machine-learning#cloudflare