RobbieJ/granite-3.1-8b-bfrpg-osft

Update model card: eval scores from qwen3-14b judge (delta -0.14)

Replace v1 with OSFT v2 (improved training data, 68% gap closure)

Add model card with training details, usage, and evaluation context

OSFT fine-tuned Granite 3.1 8B on Basic Fantasy RPG data (3 epochs, rank ratio 0.15)

initial commit