The three arrays have to be the same length: one set of logits (and one label) per input. The classic mistake is passing two inputs and two labels but only one row of logits, e.g. inputs = [text1, text2] with logits = [[0.1, 0.9]] and labels = [1, 0]. Fix it by giving logits one row per input: logits = [[0.1, 0.9], [0.8, 0.2]]. Same rule applies to any extra arrays you log alongside them; batch size must agree across all of them.

Context: Calling log_model_outputs() throws "ValueError: Inputs and logits must have the same batch size." I passed inputs, logits, and labels. What is mismatched?