Fix tests

src/diffusers/loaders/lora.py

@@ -1038,7 +1038,6 @@ def set_adapters(
         adapter_names: Union[List[str], str],
         adapter_weights: Optional[Union[float, Dict, List[float], List[Dict]]] = None,
     ):
-
         adapter_names = [adapter_names] if isinstance(adapter_names, str) else adapter_names
 
         # Expand weights into a list, one entry per adapter
@@ -1059,15 +1058,19 @@ def set_adapters(
                 text_encoder_weight = weights.pop("text_encoder", None)
                 text_encoder_2_weight = weights.pop("text_encoder_2", None)
 
-                if len(weights) >0:
-                    raise ValueError(f"Got invalid key '{weights.keys()}' in lora weight dict for adapter {adapter_name}.")
+                if len(weights) > 0:
+                    raise ValueError(
+                        f"Got invalid key '{weights.keys()}' in lora weight dict for adapter {adapter_name}."
+                    )
 
-                if  text_encoder_2_weight is not None and not hasattr(self, "text_encoder_2"):
-                    logger.warning("Lora weight dict contains text_encoder_2 weights but will be ignored because pipeline does not have text_encoder_2.")
+                if text_encoder_2_weight is not None and not hasattr(self, "text_encoder_2"):
+                    logger.warning(
+                        "Lora weight dict contains text_encoder_2 weights but will be ignored because pipeline does not have text_encoder_2."
+                    )
             else:
-               unet_weight = weights
-               text_encoder_weight = weights
-               text_encoder_2_weight = weights
+                unet_weight = weights
+                text_encoder_weight = weights
+                text_encoder_2_weight = weights
 
             unet_weights.append(unet_weight)
             text_encoder_weights.append(text_encoder_weight)

src/diffusers/loaders/unet.py

@@ -11,6 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import copy
 import inspect
 import os
 from collections import defaultdict
@@ -563,7 +564,10 @@ def _unfuse_lora_apply(self, module):
                 module.unmerge()
 
     def _expand_lora_scales_dict(
-        self, scales: Union[float, Dict], blocks_with_transformer: Dict[str, int], transformer_per_block: Dict[str, int]
+        self,
+        scales: Union[float, Dict],
+        blocks_with_transformer: Dict[str, int],
+        transformer_per_block: Dict[str, int],
     ):
         """
         Expands the inputs into a more granular dictionary. See the example below for more details.
@@ -614,6 +618,8 @@ def _expand_lora_scales_dict(
         if sorted(transformer_per_block.keys()) != ["down", "up"]:
             raise ValueError("transformer_per_block needs to be a dict with keys `'down' and `'up'`")
 
+        scales = copy.deepcopy(scales)
+
         if not isinstance(scales, dict):
             scales = {o: scales for o in ["down", "mid", "up"]}
 
@@ -631,7 +637,7 @@ def _expand_lora_scales_dict(
             # eg {"down": "block_1": 1}} to {"down": "block_1": [1, 1]}}
             for i in blocks_with_transformer[updown]:
                 block = f"block_{i}"
-                if not isinstance(scales[updown][block], dict):
+                if not isinstance(scales[updown][block], list):
                     scales[updown][block] = [scales[updown][block] for _ in range(transformer_per_block[updown])]
 
             # eg {"down": "block_1": [1, 1]}}  to {"down.block_1.0": 1, "down.block_1.1": 1}

tests/lora/test_lora_layers_peft.py

@@ -847,15 +847,15 @@ def test_simple_inference_with_text_unet_block_scale(self):
                 )
 
             weights_1 = {
-                "unet" : {
+                "unet": {
                     "down": 5,
                 }
             }
             pipe.set_adapters("adapter-1", weights_1)
             output_weights_1 = pipe(**inputs, generator=torch.manual_seed(0)).images
 
             weights_2 = {
-                "unet" : {
+                "unet": {
                     "up": 5,
                 }
             }
@@ -915,17 +915,8 @@ def test_simple_inference_with_text_unet_multi_adapter_block_lora(self):
                     self.check_if_lora_correctly_set(pipe.text_encoder_2), "Lora not correctly set in text encoder 2"
                 )
 
-            scales_1 = {
-                "unet" : {
-                    "down": 5
-                }
-            }
-            scales_2 = {
-                "unet" : {
-                    "down": 5,
-                    "mid": 5
-                }
-            }
+            scales_1 = {"unet": {"down": 5}}
+            scales_2 = {"unet": {"down": 5, "mid": 5}}
             pipe.set_adapters("adapter-1", scales_1)
 
             output_adapter_1 = pipe(**inputs, generator=torch.manual_seed(0)).images
@@ -990,7 +981,7 @@ def updown_options(blocks_with_tf, layers_per_block, value):
         def all_possible_dict_opts(unet, value):
             """
             Generate every possible combination for how a lora weight dict can be.
-            E.g. 2, {"down": 2}, {"down": [2,2,2]}, {"mid": 2, "up": [2,2,2]}, ...
+            E.g. 2, {"unet: {"down": 2}}, {"unet: {"down": [2,2,2]}}, {"unet: {"mid": 2, "up": [2,2,2]}}, ...
             """
 
             down_blocks_with_tf = [i for i, d in enumerate(unet.down_blocks) if hasattr(d, "attentions")]