I fine-tuned a 1.7B model on data that doesn't exist - a 17-minute recipe

The full methodology behind localscrub's stage-2 specialist: synthetic training data with exact...