Fig. 1: The general LLM-Prop framework.
From: LLM-Prop: predicting the properties of crystalline materials using large language models
a depicts how LLM-Prop works compared to GNNs-based models. On the left most of the figure, we show how we get the description from the crystal structure using Robocrystallographer. The middle part shows the comparison between our approach (text-based) and baselines (structure-based). The yellow colored information is related to Sodium (Na), green colored information is related to Chlorine (Cl), and blue colored information is related to additional information that text data provide such as space groups and bond distances. On the right most part of the figure, we then show our proposed LLM-Prop architecture (see 2.1 for more details). b details the T5 encoder architecture used in LLM-Prop, which is a stack of six layers of Transformer encoders47.
