rafiakedir's picture
feat: upload actual trained LoRA adapter (Qwen2.5-1.5B ORPO, 3 epochs, 36 steps)
f02a80f verified