Supporting Repository for Reproducibility StudyThis repository provides comprehensive resources for reaction prediction model reproduction studies:Preprocessed External Validation Datasets: Including literature-derived reaction sets (Sulfoxonium Ylide, Meta-C–H Functionalization, Amide Coupling). Fine-Tuned Model Checkpoints: Task-specific models (yield prediction, selectivity analysis, forward/retrosynthetic prediction) trained using the original RXNGraphormer framework, containing weights and configuration files.Training Logs & Evaluation Outputs: Complete experimental records ensuring full reproducibility.Evaluation Datasets in Test_Data Folder1. Sulfoxonium Ylide DatasetReference: Lin et al., Sci. China Chem. 68, 679–686 (2025)Description: Contains high-throughput experimentation (HTE) data for Ru-catalyzed P(O)O–H insertion reactions of sulfoxonium ylides. Features:Broad substrate scope and reaction condition coverageWell-characterized α-phosphoryloxy ketone productsExperimentally validated yield measurements2. Meta-C–H Functionalization DatasetReference: Selective functionalization of hindered meta-C–H bond of o-alkylaryl ketones, Chem (2022)Description: Automated synthesis-derived dataset for meta-selective C–H functionalization of o-alkylaryl ketones. Key characteristics:Sterically hindered substratesDiverse coupling partnersChallenging selectivity prediction benchmark3. Amide Coupling Reaction HTE DatasetReference: Zhang et al., ChemRxiv (2025)Description: Machine-guided HTE dataset of amide couplings with:High-precision yield measurementsUnbiased sampling strategy (minimal selection bias)Enhanced chemical space coverage4. Amide Coupling Literature DatasetDescription: Systematically curated collection of recent (past three years) amide coupling reactions from peer-reviewed literature. Features:Comprehensive reagent system coverageDiverse functional groups and substrate architecturesIndependent benchmark for model generalization capability