Adding support for safetensors and LoRa. (#2448)

* Adding support for `safetensors` and LoRa.

* Adding metadata.

Author: Narsil

src/diffusers/loaders.py

@@ -19,13 +19,18 @@
 
 from .models.cross_attention import LoRACrossAttnProcessor
 from .models.modeling_utils import _get_model_file
-from .utils import DIFFUSERS_CACHE, HF_HUB_OFFLINE, logging
+from .utils import DIFFUSERS_CACHE, HF_HUB_OFFLINE, is_safetensors_available, logging
+
+
+if is_safetensors_available():
+    import safetensors
 
 
 logger = logging.get_logger(__name__)
 
 
 LORA_WEIGHT_NAME = "pytorch_lora_weights.bin"
+LORA_WEIGHT_NAME_SAFE = "pytorch_lora_weights.safetensors"
 
 
 class AttnProcsLayers(torch.nn.Module):
@@ -136,28 +141,53 @@ def load_attn_procs(self, pretrained_model_name_or_path_or_dict: Union[str, Dict
         use_auth_token = kwargs.pop("use_auth_token", None)
         revision = kwargs.pop("revision", None)
         subfolder = kwargs.pop("subfolder", None)
-        weight_name = kwargs.pop("weight_name", LORA_WEIGHT_NAME)
+        weight_name = kwargs.pop("weight_name", None)
 
         user_agent = {
             "file_type": "attn_procs_weights",
             "framework": "pytorch",
         }
 
+        model_file = None
         if not isinstance(pretrained_model_name_or_path_or_dict, dict):
-            model_file = _get_model_file(
-                pretrained_model_name_or_path_or_dict,
-                weights_name=weight_name,
-                cache_dir=cache_dir,
-                force_download=force_download,
-                resume_download=resume_download,
-                proxies=proxies,
-                local_files_only=local_files_only,
-                use_auth_token=use_auth_token,
-                revision=revision,
-                subfolder=subfolder,
-                user_agent=user_agent,
-            )
-            state_dict = torch.load(model_file, map_location="cpu")
+            if is_safetensors_available():
+                if weight_name is None:
+                    weight_name = LORA_WEIGHT_NAME_SAFE
+                try:
+                    model_file = _get_model_file(
+                        pretrained_model_name_or_path_or_dict,
+                        weights_name=weight_name,
+                        cache_dir=cache_dir,
+                        force_download=force_download,
+                        resume_download=resume_download,
+                        proxies=proxies,
+                        local_files_only=local_files_only,
+                        use_auth_token=use_auth_token,
+                        revision=revision,
+                        subfolder=subfolder,
+                        user_agent=user_agent,
+                    )
+                    state_dict = safetensors.torch.load_file(model_file, device="cpu")
+                except EnvironmentError:
+                    if weight_name == LORA_WEIGHT_NAME_SAFE:
+                        weight_name = None
+            if model_file is None:
+                if weight_name is None:
+                    weight_name = LORA_WEIGHT_NAME
+                model_file = _get_model_file(
+                    pretrained_model_name_or_path_or_dict,
+                    weights_name=weight_name,
+                    cache_dir=cache_dir,
+                    force_download=force_download,
+                    resume_download=resume_download,
+                    proxies=proxies,
+                    local_files_only=local_files_only,
+                    use_auth_token=use_auth_token,
+                    revision=revision,
+                    subfolder=subfolder,
+                    user_agent=user_agent,
+                )
+                state_dict = torch.load(model_file, map_location="cpu")
         else:
             state_dict = pretrained_model_name_or_path_or_dict
 
@@ -195,8 +225,9 @@ def save_attn_procs(
         self,
         save_directory: Union[str, os.PathLike],
         is_main_process: bool = True,
-        weights_name: str = LORA_WEIGHT_NAME,
+        weights_name: str = None,
         save_function: Callable = None,
+        safe_serialization: bool = False,
     ):
         r"""
         Save an attention processor to a directory, so that it can be re-loaded using the
@@ -219,7 +250,13 @@ def save_attn_procs(
             return
 
         if save_function is None:
-            save_function = torch.save
+            if safe_serialization:
+
+                def save_function(weights, filename):
+                    return safetensors.torch.save_file(weights, filename, metadata={"format": "pt"})
+
+            else:
+                save_function = torch.save
 
         os.makedirs(save_directory, exist_ok=True)
 
@@ -237,6 +274,12 @@ def save_attn_procs(
             if filename.startswith(weights_no_suffix) and os.path.isfile(full_filename) and is_main_process:
                 os.remove(full_filename)
 
+        if weights_name is None:
+            if safe_serialization:
+                weights_name = LORA_WEIGHT_NAME_SAFE
+            else:
+                weights_name = LORA_WEIGHT_NAME
+
         # Save the model
         save_function(state_dict, os.path.join(save_directory, weights_name))
 

tests/models/test_models_unet_2d_condition.py

@@ -14,6 +14,7 @@
 # limitations under the License.
 
 import gc
+import os
 import tempfile
 import unittest
 
@@ -372,6 +373,65 @@ def test_lora_save_load(self):
 
         with tempfile.TemporaryDirectory() as tmpdirname:
             model.save_attn_procs(tmpdirname)
+            self.assertTrue(os.path.isfile(os.path.join(tmpdirname, "pytorch_lora_weights.bin")))
+            torch.manual_seed(0)
+            new_model = self.model_class(**init_dict)
+            new_model.to(torch_device)
+            new_model.load_attn_procs(tmpdirname)
+
+        with torch.no_grad():
+            new_sample = new_model(**inputs_dict, cross_attention_kwargs={"scale": 0.5}).sample
+
+        assert (sample - new_sample).abs().max() < 1e-4
+
+        # LoRA and no LoRA should NOT be the same
+        assert (sample - old_sample).abs().max() > 1e-4
+
+    def test_lora_save_load_safetensors(self):
+        # enable deterministic behavior for gradient checkpointing
+        init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
+
+        init_dict["attention_head_dim"] = (8, 16)
+
+        torch.manual_seed(0)
+        model = self.model_class(**init_dict)
+        model.to(torch_device)
+
+        with torch.no_grad():
+            old_sample = model(**inputs_dict).sample
+
+        lora_attn_procs = {}
+        for name in model.attn_processors.keys():
+            cross_attention_dim = None if name.endswith("attn1.processor") else model.config.cross_attention_dim
+            if name.startswith("mid_block"):
+                hidden_size = model.config.block_out_channels[-1]
+            elif name.startswith("up_blocks"):
+                block_id = int(name[len("up_blocks.")])
+                hidden_size = list(reversed(model.config.block_out_channels))[block_id]
+            elif name.startswith("down_blocks"):
+                block_id = int(name[len("down_blocks.")])
+                hidden_size = model.config.block_out_channels[block_id]
+
+            lora_attn_procs[name] = LoRACrossAttnProcessor(
+                hidden_size=hidden_size, cross_attention_dim=cross_attention_dim
+            )
+            lora_attn_procs[name] = lora_attn_procs[name].to(model.device)
+
+            # add 1 to weights to mock trained weights
+            with torch.no_grad():
+                lora_attn_procs[name].to_q_lora.up.weight += 1
+                lora_attn_procs[name].to_k_lora.up.weight += 1
+                lora_attn_procs[name].to_v_lora.up.weight += 1
+                lora_attn_procs[name].to_out_lora.up.weight += 1
+
+        model.set_attn_processor(lora_attn_procs)
+
+        with torch.no_grad():
+            sample = model(**inputs_dict, cross_attention_kwargs={"scale": 0.5}).sample
+
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            model.save_attn_procs(tmpdirname, safe_serialization=True)
+            self.assertTrue(os.path.isfile(os.path.join(tmpdirname, "pytorch_lora_weights.safetensors")))
             torch.manual_seed(0)
             new_model = self.model_class(**init_dict)
             new_model.to(torch_device)