Fix a couple of bugs to get basic vanilla partial model load working with the model cache.

invokeai/backend/model_manager/load/load_base.py

@@ -75,7 +75,7 @@ class LoadedModelWithoutConfig:
     @property
     def model(self) -> AnyModel:
         """Return the model without locking it."""
-        return self._cache_record.model
+        return self._cache_record.cached_model.model
 
 
 class LoadedModel(LoadedModelWithoutConfig):

invokeai/backend/model_manager/load/model_cache/model_cache.py

@@ -219,9 +219,11 @@ class ModelCache:
             # self._print_cuda_stats()
         except torch.cuda.OutOfMemoryError:
             self._logger.warning("Insufficient GPU memory to load model. Aborting")
-            raise
-        finally:
             cache_entry.unlock()
+            raise
+        except Exception:
+            cache_entry.unlock()
+            raise
 
         # try:
         #     if self._lazy_offloading: