[VLM][Core] Support profiling with multiple multi-modal inputs per prompt (#7126)

docs/source/models/enabling_multimodal_inputs.rst

@@ -66,7 +66,7 @@ A default mapper is available for each modality in the core vLLM library. This i
 3. Register maximum number of multi-modal tokens
 ------------------------------------------------
 
-For each modality type that the model accepts as input, calculate the maximum possible number of tokens
+For each modality type that the model accepts as input, calculate the maximum possible number of tokens per data instance
 and register it via :meth:`INPUT_REGISTRY.register_dummy_data <vllm.inputs.registry.InputRegistry.register_max_multimodal_tokens>`.
 
 .. code-block:: diff