Fig. 2: Input prompt design for LLM-based OSC property prediction.
From: From Corpus to Innovation: Advancing Organic Solar Cell Design with Large Language Models
Prompt for predicting power conversion efficiency (PCE). smiles_donor and smiles_acceptor are SMILES strings for donor and acceptor materials, respectively. blend_ratio denotes the ratio between donor and acceptor in the active layer. solvent specifies the solvent name, including percentages if multiple solvents or additives are present. processing_technique indicates the thin-film preparation method, such as “spin coating”. For thermal annealing (TA) or solvent vapor annealing (SVA), if True, provide the condition, for example “TA at 150°C for 10 minutes”. thickness_nm represents the active layer’s thickness in nm, while molecular_weight is the polymer’s molecular weight in g/mol. If information for blend_ratio, solvent, processing_technique, TA, SVA, thickness_nm, or molecular_weight is unavailable, it is marked as unknown. During fine-tuning, the PCE value is provided but omitted during prediction. < s > and < /s > denote start and end tokens. For JSC, VOC, and FF, similar prompts are used, with PCE replaced by the corresponding property.
