diff --git a/config/example.yaml b/config/example.yaml
index 449b0c1..8667448 100644
--- a/config/example.yaml
+++ b/config/example.yaml
@@ -23,3 +23,5 @@ download_max_set_size: 50
 data_loader_num_workers: 4
 models_dag_config_name: dspn_deepnarrow
 models_dag_days_interval: 2
+
+trained_models_dirpath: trained_models/
diff --git a/models/generic_model.py b/models/generic_model.py
index 3d758e6..cd27fc6 100644
--- a/models/generic_model.py
+++ b/models/generic_model.py
@@ -1,6 +1,13 @@
 import pytorch_lightning as pl
 import torch
+import os
+import sys
+import pickle
 
+project_root_dir = os.path.abspath(os.path.dirname(os.path.abspath(__file__))+"/..")
+sys.path.insert(0, project_root_dir)
+
+from common import config
 
 class AEModule(torch.nn.Module):
     """
@@ -14,9 +21,6 @@ def __init__(self,**kwargs):
         self.rel = kwargs.get('rel', None)
         super().__init__()
 
-    @property
-    def name(self):
-        return f"{self.classname}_{self.rel.name}"
 
     def activate(self, x):
         return self.activation_function()(x)
@@ -152,7 +156,18 @@ def classname(self):
 
     @property
     def name(self):
-        raise Exception("implement in subclass")
+        return f"{self.classname}_{self.rel.name}"
+
+    @property
+    def kwargs_filename(self):
+        os.path.join(
+            config.trained_models_dirpath,
+            f"{self.name}_kwargs.pickle"
+        )
+
+    def dump_kwargs(self):
+        with open(self.kwargs_filename, 'wb') as handle:
+            pickle.dump(self.kwargs, handle, protocol=pickle.HIGHEST_PROTOCOL)
 
     def make_train_loader(self,tsets):
         raise Exception("implement in subclass")
diff --git a/models/models_storage.py b/models/models_storage.py
index ee1572a..7958d47 100644
--- a/models/models_storage.py
+++ b/models/models_storage.py
@@ -1,4 +1,10 @@
+import logging
 import os
+import sys
+import pickle
+
+project_root_dir = os.path.abspath(os.path.dirname(os.path.abspath(__file__))+"/..")
+sys.path.insert(0, project_root_dir)
 
 from common import utils, relspecs
 from models import dspn_autoencoder
@@ -10,11 +16,35 @@ class DSPNAEModelsStorage(utils.Collection): # FIXME: abstraction?
     the models.
     """
     def __init__(self):
-
+        os.chdir(project_root_dir)
         for rel in relspecs.rels:
-            model_filename = os.path.join(
-                "trained_models",
-                f"DSPNAE_{rel.name}"
-            )
-            model = dspn_autoencoder.DSPNAE.load_from_checkpoint(model_filename)
+            model = self.load(rel)
             self[rel.name] = model
+
+    def load(self, rel):
+        model_filename = os.path.join(
+            "trained_models",
+            f"DSPNAE_{rel.name}.ckpt"
+        )
+
+        # FIXME: duplicated from GenericModel
+        kwargs_filename = f"DSPNAE_{rel.name}_kwargs.pickle"
+        with open(kwargs_filename) as f:
+            kwargs = pickle.load(f)
+        logging.info(f"loading {model_filename}..")
+        if not os.path.exists(model_filename):
+            logging.warning(f"could not find model saved in {model_filename}")
+            return
+
+        model = dspn_autoencoder.DSPNAE(**kwargs)
+        model.load_from_checkpoint(model_filename)
+        return model
+
+
+def test():
+    ms = DSPNAEModelsStorage()
+    print("ms",ms)
+    print("done.")
+
+if __name__ == "__main__":
+    test()
\ No newline at end of file
diff --git a/models/run.py b/models/run.py
index dee648e..318a786 100644
--- a/models/run.py
+++ b/models/run.py
@@ -6,8 +6,8 @@
 import os
 import sys
 import argparse
-
-from common import utils, relspecs, persistency
+import pickle
+from common import utils, relspecs, persistency, config
 from models import diagnostics, measurements as ms
 
 def get_args():
@@ -153,13 +153,13 @@ def run(Model,config_name, dynamic_config={}):
             item_dim=tsets.item_dim,
             **model_config
         ).to(device)
-
+        model.dump_kwargs()
         train_loader = model.make_train_loader(tsets)
         test_loader = model.make_test_loader(tsets)
         model_filename = model.name
         checkpoint_callback = pl.callbacks.ModelCheckpoint(
             monitor="train_loss",
-            dirpath="trained_models/",
+            dirpath=config.trained_models_dirpath,
             filename=model_filename,
             save_top_k=1,
             mode="min",