RxnNano:Training Compact LLMs for Chemical Reaction and Retrosynthesis Prediction via Hierarchical Curriculum Learning

Researchers developed RxnNano, a compact 0.5B-parameter AI model for chemical reaction prediction that outperforms much larger 7B+ parameter models by 23.5% through novel training techniques focused on chemical understanding rather than scale. The framework uses hierarchical curriculum learning and chemical consistency objectives to improve drug discovery and synthesis planning applications.

• →RxnNano achieves 23.5% better Top-1 accuracy than models 10x larger through chemical understanding rather than parameter scaling
• →The model uses Latent Chemical Consistency to ensure physically plausible and reversible chemical transformations
• →Hierarchical Cognitive Curriculum trains the model progressively from syntax mastery to semantic chemical reasoning
• →Atom-Map Permutation Invariance forces the model to learn invariant relational topology for better chemical intuition
• →The compact 0.5B parameter model demonstrates that efficiency and specialized training can outperform brute-force scaling approaches