Añadir nuevas dependencias al instalador

Se añaden dos nuevas instalaciones de paquetes en el archivo `app.py`:
- Se instala `flash-attn` sin aislamiento de construcción.
- Se agrega la instalación de `bpy` versión 3.6.0 desde el índice de Blender.

Archivo modificado: [app.py](app.py)

app.py

@@ -25,6 +25,8 @@ else:
 
 subprocess.run(f'pip install spconv{spconv_version}', shell=True)
 subprocess.run(f'pip install torch_scatter torch_cluster -f https://data.pyg.org/whl/torch-{torch_version}+{cuda_version}.html --no-cache-dir', shell=True)
+subprocess.run(f'pip install flash-attn --no-build-isolation --no-cache-dir', shell=True)
+subprocess.run(f'pip install bpy==3.6.0 --extra-index-url https://download.blender.org/pypi/', shell=True)
 
 # Helper functions
 def validate_input_file(file_path: str, supported_formats: list) -> bool:
@@ -72,6 +74,7 @@ def extract_mesh_python(input_file: str, output_dir: str) -> str:
 
     return expected_npz_dir
 
+
 def run_inference_python(
         input_file: str,
         output_file: str,
@@ -199,31 +202,24 @@ def run_inference_python(
 
     callbacks.append(get_writer(**writer_config, order_config=predict_transform_config.order_config))
 
-    # Add ModelCheckpoint callback if present in task callbacks to avoid Lightning warning
     checkpoint_callbacks = []
     if hasattr(task, 'callbacks') and task.callbacks:
         for cb in task.callbacks:
             if isinstance(cb, dict) and cb.get('__target__', '').startswith('ModelCheckpoint'):
-                # Remove __target__ key and pass rest as kwargs
                 cb_kwargs = {k: v for k, v in cb.items() if k != '__target__'}
                 checkpoint_callbacks.append(ModelCheckpoint(**cb_kwargs))
-    callbacks = checkpoint_callbacks + callbacks
 
-    # Get system
-    if system_config_path is None or not Path(system_config_path).exists():
-        raise FileNotFoundError(f"System configuration file not found: {system_config_path}")
-    system_config = Box(yaml.safe_load(open(system_config_path, 'r')))
-    system = get_system(**system_config, model=model, steps_per_epoch=1)
+    # Agregar callbacks writer y checkpoint juntos
+    callbacks = checkpoint_callbacks + [get_writer(**writer_config, order_config=predict_transform_config.order_config)]
 
-    # Setup trainer
     trainer_config = task.trainer
     resume_from_checkpoint = download(task.resume_from_checkpoint)
 
     trainer = L.Trainer(callbacks=callbacks, logger=None, **trainer_config)
 
-    # Run prediction
     trainer.predict(system, datamodule=data, ckpt_path=resume_from_checkpoint, return_predictions=False)
 
+
     # Handle output file location and validation
     if inference_type == "skeleton":
         input_name_stem = Path(input_file).stem