THD Model Eval — RAFT v6

LoRA fine-tuned vs base model output comparison across 10 evaluation cases (4 generations each)

raft_v6 eval_lora eval_base 10 cases × 4 gens