Valid on initial model weights

allenact/algorithms/onpolicy_sync/engine.py

@@ -1312,7 +1312,7 @@ def _save_checkpoint_then_send_checkpoint_for_validation_and_update_last_save_co
                 self.checkpoints_queue.put(("eval", model_path))
         self.last_save = self.training_pipeline.total_steps
 
-    def run_pipeline(self):
+    def run_pipeline(self, valid_on_initial_weights: bool = False):
         cur_stage_training_settings = (
             self.training_pipeline.current_stage.training_settings
         )
@@ -1336,6 +1336,16 @@ def run_pipeline(self):
         )
         already_saved_checkpoint = False
 
+        if (
+            valid_on_initial_weights
+            and should_save_checkpoints
+            and self.checkpoints_queue is not None
+        ):
+            if self.worker_id == self.first_local_worker_id:
+                model_path = self.checkpoint_save()
+                if self.checkpoints_queue is not None:
+                    self.checkpoints_queue.put(("eval", model_path))
+
         while True:
             pipeline_stage_changed = self.training_pipeline.before_rollout(
                 train_metrics=self._last_aggregated_train_task_metrics
@@ -1572,7 +1582,10 @@ def run_pipeline(self):
                 )
 
     def train(
-        self, checkpoint_file_name: Optional[str] = None, restart_pipeline: bool = False
+        self,
+        checkpoint_file_name: Optional[str] = None,
+        restart_pipeline: bool = False,
+        valid_on_initial_weights: bool = False,
     ):
         assert (
             self.mode == TRAIN_MODE_STR
@@ -1584,7 +1597,7 @@ def train(
             if checkpoint_file_name is not None:
                 self.checkpoint_load(checkpoint_file_name, restart_pipeline)
 
-            self.run_pipeline()
+            self.run_pipeline(valid_on_initial_weights=valid_on_initial_weights)
 
             training_completed_successfully = True
         except KeyboardInterrupt:

allenact/algorithms/onpolicy_sync/runner.py

@@ -316,6 +316,7 @@ def train_loop(
         id: int = 0,
         checkpoint: Optional[str] = None,
         restart_pipeline: bool = False,
+        valid_on_initial_weights: bool = False,
         *engine_args,
         **engine_kwargs,
     ):
@@ -333,7 +334,9 @@ def train_loop(
         if trainer is not None:
             OnPolicyRunner.init_process("Train", id, to_close_on_termination=trainer)
             trainer.train(
-                checkpoint_file_name=checkpoint, restart_pipeline=restart_pipeline
+                checkpoint_file_name=checkpoint,
+                restart_pipeline=restart_pipeline,
+                valid_on_initial_weights=valid_on_initial_weights,
             )
 
     @staticmethod
@@ -407,6 +410,7 @@ def start_train(
         max_sampler_processes_per_worker: Optional[int] = None,
         save_ckpt_after_every_pipeline_stage: bool = True,
         collect_valid_results: bool = False,
+        valid_on_initial_weights: bool = False,
     ):
         self._initialize_start_train_or_start_test()
 
@@ -457,6 +461,7 @@ def start_train(
                 if model_hash is None
                 else model_hash,
                 first_local_worker_id=worker_ids[0],
+                valid_on_initial_weights=valid_on_initial_weights,
             )
             train: BaseProcess = self.mp_ctx.Process(
                 target=self.train_loop, kwargs=training_kwargs,

allenact/main.py

@@ -220,6 +220,15 @@ def get_argument_parser():
     )
     parser.set_defaults(collect_valid_results=False)
 
+    parser.add_argument(
+        "--valid_on_initial_weights",
+        dest="valid_on_initial_weights",
+        action="store_true",
+        required=False,
+        help="enables running validation on the model with initial weights",
+    )
+    parser.set_defaults(collect_valid_results=False)
+
     parser.add_argument(
         "--test_expert",
         dest="test_expert",
@@ -443,6 +452,7 @@ def main():
             restart_pipeline=args.restart_pipeline,
             max_sampler_processes_per_worker=args.max_sampler_processes_per_worker,
             collect_valid_results=args.collect_valid_results,
+            valid_on_initial_weights=args.valid_on_initial_weights,
         )
     else:
         OnPolicyRunner(