LoRA model loading fixes (#3663)

This PR enables model manager importation of diffusers-style .bin LoRAs.
However, since there is no backend support for this type of LoRA yet,
attempts to use them will result in an unimplemented error.

It closes #3636 and #3637

Author: blessedcoolant

invokeai/backend/install/model_install_backend.py

@@ -193,7 +193,10 @@ def heuristic_import(self,
                 models_installed.update(self._install_path(path))
 
             # folders style or similar
-            elif path.is_dir() and any([(path/x).exists() for x in {'config.json','model_index.json','learned_embeds.bin'}]):
+            elif path.is_dir() and any([(path/x).exists() for x in \
+                                        {'config.json','model_index.json','learned_embeds.bin','pytorch_lora_weights.bin'}
+                                        ]
+                                       ):
                 models_installed.update(self._install_path(path))
 
             # recursive scan

invokeai/backend/model_management/model_manager.py

@@ -785,7 +785,7 @@ def autoimport(self)->Dict[str, AddModelResult]:
                     if path in known_paths or path.parent in scanned_dirs:
                         scanned_dirs.add(path)
                         continue
-                    if any([(path/x).exists() for x in {'config.json','model_index.json','learned_embeds.bin'}]):
+                    if any([(path/x).exists() for x in {'config.json','model_index.json','learned_embeds.bin','pytorch_lora_weights.bin'}]):
                         new_models_found.update(installer.heuristic_import(path))
                         scanned_dirs.add(path)
 
@@ -794,7 +794,8 @@ def autoimport(self)->Dict[str, AddModelResult]:
                     if path in known_paths or path.parent in scanned_dirs:
                         continue
                     if path.suffix in {'.ckpt','.bin','.pth','.safetensors','.pt'}:
-                        new_models_found.update(installer.heuristic_import(path))
+                        import_result = installer.heuristic_import(path)
+                        new_models_found.update(import_result)
 
             self.logger.info(f'Scanned {items_scanned} files and directories, imported {len(new_models_found)} models')
             installed.update(new_models_found)

invokeai/backend/model_management/model_probe.py

@@ -78,7 +78,6 @@ def probe(cls,
             format_type = 'diffusers' if model_path.is_dir() else 'checkpoint'
         else:
             format_type = 'diffusers' if isinstance(model,(ConfigMixin,ModelMixin)) else 'checkpoint'
-
         model_info = None
         try:
             model_type = cls.get_model_type_from_folder(model_path, model) \
@@ -105,7 +104,7 @@ def probe(cls,
                                      ) else 512,
             )
         except Exception:
-            return None
+            raise
 
         return model_info
 
@@ -127,6 +126,8 @@ def get_model_type_from_checkpoint(cls, model_path: Path, checkpoint: dict) -> M
                 return ModelType.Vae
             elif any(key.startswith(v) for v in {"lora_te_", "lora_unet_"}):
                 return ModelType.Lora
+            elif any(key.endswith(v) for v in {"to_k_lora.up.weight", "to_q_lora.down.weight"}):
+                return ModelType.Lora
             elif any(key.startswith(v) for v in {"control_model", "input_blocks"}):
                 return ModelType.ControlNet
             elif key in {"emb_params", "string_to_param"}:
@@ -137,7 +138,7 @@ def get_model_type_from_checkpoint(cls, model_path: Path, checkpoint: dict) -> M
             if len(ckpt) < 10 and all(isinstance(v, torch.Tensor) for v in ckpt.values()):
                 return ModelType.TextualInversion
 
-        raise ValueError("Unable to determine model type")
+        raise ValueError(f"Unable to determine model type for {model_path}")
 
     @classmethod
     def get_model_type_from_folder(cls, folder_path: Path, model: ModelMixin)->ModelType:
@@ -167,7 +168,7 @@ def get_model_type_from_folder(cls, folder_path: Path, model: ModelMixin)->Model
             return type
 
         # give up
-        raise ValueError("Unable to determine model type")
+        raise ValueError("Unable to determine model type for {folder_path}")
 
     @classmethod
     def _scan_and_load_checkpoint(cls,model_path: Path)->dict:

invokeai/frontend/install/model_install.py

@@ -678,9 +678,8 @@ def select_and_download_models(opt: Namespace):
 
     # this is where the TUI is called
     else:
-        # needed because the torch library is loaded, even though we don't use it
-        # currently commented out because it has started generating errors (?)
-        # torch.multiprocessing.set_start_method("spawn")
+        # needed to support the probe() method running under a subprocess
+        torch.multiprocessing.set_start_method("spawn")
 
         # the third argument is needed in the Windows 11 environment in
         # order to launch and resize a console window running this program