Visualize predicted turn endings from two models.Red ⇒ agent should reply • Blue ⇒ agent should wait
Input text should follow the following chat template. Transcripts should be normalized to be lowercase and without punctuation.