refactor(pipeline): Enhance state filename generation and profiling method

- Updated state filename generation to use the registry name when available, improving clarity in saved files.
- Modified the profile_steps method to include a warmup_runs parameter, allowing for more controlled performance profiling.
- Ensured consistent conditions during profiling by deep copying transitions for each run, enhancing accuracy in timing results.

src/lerobot/processor/pipeline.py

@@ -448,7 +448,12 @@ class RobotProcessor(ModelHubMixin):
                     for key, tensor in state.items():
                         cloned_state[key] = tensor.clone()
 
-                    state_filename = f"step_{step_index}.safetensors"
+                    # Use registry name for more meaningful filenames when available
+                    if registry_name:
+                        state_filename = f"{registry_name}.safetensors"
+                    else:
+                        state_filename = f"step_{step_index}.safetensors"
+
                     save_file(cloned_state, os.path.join(destination_path, state_filename))
                     step_entry["state_file"] = state_filename
 
@@ -707,23 +712,37 @@ class RobotProcessor(ModelHubMixin):
         for fn in self.reset_hooks:
             fn()
 
-    def profile_steps(self, transition: EnvTransition, num_runs: int = 100) -> dict[str, float]:
+    def profile_steps(
+        self, transition: EnvTransition, num_runs: int = 100, warmup_runs: int = 5
+    ) -> dict[str, float]:
         """Profile the execution time of each step for performance optimization."""
+        import copy
         import time
 
         profile_results = {}
 
+        # Make a copy to avoid altering the original transition
+        transition_copy = copy.deepcopy(transition)
+
+        # Get intermediate transitions for each step using step_through
+        intermediate_transitions = list(self.step_through(transition_copy))
+
         for idx, processor_step in enumerate(self.steps):
             step_name = f"step_{idx}_{processor_step.__class__.__name__}"
 
-            # Warm up
+            # Use the appropriate input transition for this step
-            for _ in range(5):
+            input_transition = intermediate_transitions[idx]
-                _ = processor_step(transition)
 
-            # Time the step
+            # Warm up - copy transition for each run to ensure consistent conditions
+            for _ in range(warmup_runs):
+                transition_copy = copy.deepcopy(input_transition)
+                _ = processor_step(transition_copy)
+
+            # Time the step - copy transition for each run to ensure consistent conditions
             start_time = time.perf_counter()
             for _ in range(num_runs):
-                transition = processor_step(transition)
+                transition_copy = copy.deepcopy(input_transition)
+                _ = processor_step(transition_copy)
             end_time = time.perf_counter()
 
             avg_time = (end_time - start_time) / num_runs * 1000  # Convert to milliseconds