Make --torch use torch even without config (#4400)

* Make --torch use torch even without config

* Update ml-agents/mlagents/trainers/trainer_util.py

Co-authored-by: andrewcoh <54679309+andrewcoh@users.noreply.github.com>

* Update ml-agents/mlagents/trainers/trainer_util.py

Co-authored-by: andrewcoh <54679309+andrewcoh@users.noreply.github.com>

* Update ml-agents/mlagents/trainers/trainer_util.py

Co-authored-by: andrewcoh <54679309+andrewcoh@users.noreply.github.com>

* renaming use_torch to force_torch

Co-authored-by: andrewcoh <54679309+andrewcoh@users.noreply.github.com>

ml-agents/mlagents/trainers/learn.py

     GaugeWriter,
     ConsoleWriter,
 )
-from mlagents.trainers.cli_utils import parser
+from mlagents.trainers.cli_utils import parser, DetectDefault
 from mlagents_envs.environment import UnityEnvironment
 from mlagents.trainers.settings import RunOptions
 
         )
 
         trainer_factory = TrainerFactory(
-            options.behaviors,
-            write_path,
-            not checkpoint_settings.inference,
-            checkpoint_settings.resume,
-            run_seed,
-            env_parameter_manager,
-            maybe_init_path,
-            False,
+            trainer_config=options.behaviors,
+            output_path=write_path,
+            train_model=not checkpoint_settings.inference,
+            load_model=checkpoint_settings.resume,
+            seed=run_seed,
+            param_manager=env_parameter_manager,
+            init_path=maybe_init_path,
+            multi_gpu=False,
+            force_torch="torch" in DetectDefault.non_default_args,
         )
         # Create controller and begin training.
         tc = TrainerController(

ml-agents/mlagents/trainers/settings.py

                 else:  # Base options
                     configured_dict[key] = val
 
-        # Apply --torch retroactively
-        if "torch" in DetectDefault.non_default_args:
-            for trainer_set in final_runoptions.behaviors.values():
-                trainer_set.framework = FrameworkType.PYTORCH
         return final_runoptions
 
     @staticmethod

ml-agents/mlagents/trainers/trainer_util.py

 from mlagents.trainers.sac.trainer import SACTrainer
 from mlagents.trainers.ghost.trainer import GhostTrainer
 from mlagents.trainers.ghost.controller import GhostController
-from mlagents.trainers.settings import TrainerSettings, TrainerType
+from mlagents.trainers.settings import TrainerSettings, TrainerType, FrameworkType
 
 
 logger = get_logger(__name__)
         param_manager: EnvironmentParameterManager,
         init_path: str = None,
         multi_gpu: bool = False,
+        force_torch: bool = False,
+        """
+        The TrainerFactory generates the Trainers based on the configuration passed as
+        input.
+        :param trainer_config: A dictionary from behavior name to TrainerSettings
+        :param output_path: The path to the directory where the artifacts generated by
+        the trainer will be saved.
+        :param train_model: If True, the Trainers will train the model and if False,
+        only perform inference.
+        :param load_model: If True, the Trainer will load neural networks weights from
+        the previous run.
+        :param seed: The seed of the Trainers. Dictates how the neural networks will be
+        initialized.
+        :param param_manager: The EnvironmentParameterManager that will dictate when/if
+        the EnvironmentParameters must change.
+        :param init_path: Path from which to load model.
+        :param multi_gpu: If True, multi-gpu will be used. (currently not available)
+        :param force_torch: If True, the Trainers will all use the PyTorch framework
+        instead of the TensorFlow framework.
+        """
         self.trainer_config = trainer_config
         self.output_path = output_path
         self.init_path = init_path
         self.param_manager = param_manager
         self.multi_gpu = multi_gpu
         self.ghost_controller = GhostController()
+        self._force_torch = force_torch
-    def generate(self, brain_name: str) -> Trainer:
-        if brain_name not in self.trainer_config.keys():
+    def generate(self, behavior_name: str) -> Trainer:
+        if behavior_name not in self.trainer_config.keys():
-                f"Behavior name {brain_name} does not match any behaviors specified in the trainer configuration file:"
-                f"{sorted(self.trainer_config.keys())}"
+                f"Behavior name {behavior_name} does not match any behaviors specified"
+                f"in the trainer configuration file: {sorted(self.trainer_config.keys())}"
+        trainer_settings = self.trainer_config[behavior_name]
+        if self._force_torch:
+            trainer_settings.framework = FrameworkType.PYTORCH
-            self.trainer_config[brain_name],
-            brain_name,
+            trainer_settings,
+            behavior_name,
             self.output_path,
             self.train_model,
             self.load_model,