Detecting out-of-distribution text using topological features of transformer-based language models

Open in new window