Extended Data Table 1 Example prompts and completions for predicting the phase of high-entropy alloys

From: Leveraging large language models for predictive chemistry

  1. These models have been trained using a self-supervised approach, that is, to predict the next token given an input text sequence. This implies we offer the list of questions and answers as one large string. The program learns that in our string ‘###’ indicates the end of a prompt and ‘@@@’ the end of a completion. Here, we used the fact that learning one character is cheaper and easier, hence 0=multi-phase.
Back to article page