Updated  1 month ago

The first fine-tune attempt. Trained on gemma-3-1b-it with rank 16 for 3 epochs on deterministic targets. It parroted the same settings template for every input — it memorized, it did not judge.