Fig. 3: Performance comparison of LLM-Prop with baselines on test set across different input modalities. | npj Computational Materials

Fig. 3: Performance comparison of LLM-Prop with baselines on test set across different input modalities.

From: LLM-Prop: predicting the properties of crystalline materials using large language models

Fig. 3

“Robo.” means that the input is a condensed structure in a JSON format generated by Robocrystallographer, “CIF” denotes that the input is a CIF file, and “Descr.” represents having input as crystal descriptions. For description-based models we only show the best performing input preprocessing strategy (refer to Table 2 for more details). From (a–f) represents the performance visualization when predicting Band gap, FEPA, EPA, Volume, Ehull, and Is-gap-direct, respectively.

Back to article page