paper_type.json (254B)
1 { 2 "paper_type": "empirical", 3 "reason": "Proposes type-constrained decoding for LLM functions and validates experimentally with quantitative results (7% accuracy improvement, 53% latency improvement) on established benchmarks (HybridQA, TAG-Bench)." 4 }