japhba's picture
LoRA checkpoint: Qwen3-8B fine-tuned on 500 synthetic facts (20 epochs, loss=0.0003)
7381bf5 verified