Behavior of batching when input batches exceed max_batch_size in Triton

[qocharian]

We have set max_batch_size: 32 in our Triton model configuration. After some processing inside our pipeline, the inputs have shape [B, T, D] and are passed further.
My question is: what happens if we receive two consecutive input batches of size [17, T, D] and [17, T, D]? Since 17 + 17 > 32, does Triton merge them into [32, T, D] and [2, T, D], or does it handle them separately as they are? Could you please confirm if this is expected behavior? If so, are there any recommendations on how to improve batch utilization for such cases?