Add profiling capabilities for processor components in streaming model. Introduce _patch_component_call to time component invocations and enhance _install_profiling_hooks with new processor stages. Refactor tensor conversion and device transfer functions for improved clarity and performance. Update profiling summary to include new stages and advice for optimization.

streaming_turn_model.py

@@ -565,12 +565,31 @@ def _prepare_conversation(
     ):
         processor_inputs["images"] = images
     batch = processor(**processor_inputs)
-    batch = batch.convert_to_tensors("pt")
+    batch = _processor_batch_convert_to_tensors(batch)
     # BatchFeature.to(dtype) changes floating tensors only; IDs stay integral.
-    batch = batch.to(model.device).to(model.dtype)
+    batch = _processor_batch_to_device(batch, model.device)
+    batch = _processor_batch_to_dtype(batch, model.dtype)
     return dict(batch)
 
 
+def _processor_batch_convert_to_tensors(batch: Any) -> Any:
+    """Materialize the processor's NumPy/list outputs as PyTorch tensors."""
+
+    return batch.convert_to_tensors("pt")
+
+
+def _processor_batch_to_device(batch: Any, device: torch.device) -> Any:
+    """Move one processor batch to the model device."""
+
+    return batch.to(device)
+
+
+def _processor_batch_to_dtype(batch: Any, dtype: torch.dtype) -> Any:
+    """Cast only floating processor tensors to the model dtype."""
+
+    return batch.to(dtype)
+
+
 def _readout_query_ids(
     processor: Qwen2_5OmniProcessor,
     device: torch.device,