I had Gemini train its own replacement for $9

Gemini 3.1 Pro labeled 4,290 Reddit comments for $9; a fine-tuned GLiNER model now tags brands, models and materials locally at 0.83 F1 against those labels. Five of ten training runs produced nothing because of a tensor I misread as an attention mask; the loss went flat and no error named the tensor.