Deprecate sample size (#1406)

* up

* up

* fix

* uP

* more fixes

* up

* uP

* up

* up

* uP

* fix final tests

Author: patrickvonplaten

src/diffusers/configuration_utils.py

@@ -91,9 +91,6 @@ class ConfigMixin:
     def register_to_config(self, **kwargs):
         if self.config_name is None:
             raise NotImplementedError(f"Make sure that {self.__class__} has defined a class name `config_name`")
-        kwargs["_class_name"] = self.__class__.__name__
-        kwargs["_diffusers_version"] = __version__
-
         # Special case for `kwargs` used in deprecation warning added to schedulers
         # TODO: remove this when we remove the deprecation warning, and the `kwargs` argument,
         # or solve in a more general way.
@@ -462,7 +459,7 @@ def extract_init_dict(cls, config_dict, **kwargs):
         unused_kwargs = {**config_dict, **kwargs}
 
         # 7. Define "hidden" config parameters that were saved for compatible classes
-        hidden_config_dict = {k: v for k, v in original_dict.items() if k not in init_dict and not k.startswith("_")}
+        hidden_config_dict = {k: v for k, v in original_dict.items() if k not in init_dict}
 
         return init_dict, unused_kwargs, hidden_config_dict
 
@@ -493,6 +490,9 @@ def to_json_string(self) -> str:
             `str`: String containing all the attributes that make up this configuration instance in JSON format.
         """
         config_dict = self._internal_dict if hasattr(self, "_internal_dict") else {}
+        config_dict["_class_name"] = self.__class__.__name__
+        config_dict["_diffusers_version"] = __version__
+
         return json.dumps(config_dict, indent=2, sort_keys=True) + "\n"
 
     def to_json_file(self, json_file_path: Union[str, os.PathLike]):
@@ -520,6 +520,7 @@ def register_to_config(init):
     def inner_init(self, *args, **kwargs):
         # Ignore private kwargs in the init.
         init_kwargs = {k: v for k, v in kwargs.items() if not k.startswith("_")}
+        config_init_kwargs = {k: v for k, v in kwargs.items() if k.startswith("_")}
         init(self, *args, **init_kwargs)
         if not isinstance(self, ConfigMixin):
             raise RuntimeError(
@@ -545,6 +546,7 @@ def inner_init(self, *args, **kwargs):
                 if k not in ignore and k not in new_kwargs
             }
         )
+        new_kwargs = {**config_init_kwargs, **new_kwargs}
         getattr(self, "register_to_config")(**new_kwargs)
 
     return inner_init
@@ -562,7 +564,7 @@ def init(self, *args, **kwargs):
             )
 
         # Ignore private kwargs in the init. Retrieve all passed attributes
-        init_kwargs = {k: v for k, v in kwargs.items() if not k.startswith("_")}
+        init_kwargs = {k: v for k, v in kwargs.items()}
 
         # Retrieve default values
         fields = dataclasses.fields(self)

src/diffusers/modeling_utils.py

@@ -448,7 +448,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
         if low_cpu_mem_usage:
             # Instantiate model with empty weights
             with accelerate.init_empty_weights():
-                model, unused_kwargs = cls.from_config(
+                config, unused_kwargs = cls.load_config(
                     config_path,
                     cache_dir=cache_dir,
                     return_unused_kwargs=True,
@@ -462,6 +462,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
                     device_map=device_map,
                     **kwargs,
                 )
+                model = cls.from_config(config, **unused_kwargs)
 
             # if device_map is Non,e load the state dict on move the params from meta device to the cpu
             if device_map is None:
@@ -482,7 +483,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
                 "error_msgs": [],
             }
         else:
-            model, unused_kwargs = cls.from_config(
+            config, unused_kwargs = cls.load_config(
                 config_path,
                 cache_dir=cache_dir,
                 return_unused_kwargs=True,
@@ -496,6 +497,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
                 device_map=device_map,
                 **kwargs,
             )
+            model = cls.from_config(config, **unused_kwargs)
 
             state_dict = load_state_dict(model_file)
             model, missing_keys, unexpected_keys, mismatched_keys, error_msgs = cls._load_pretrained_model(

src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py

@@ -18,6 +18,7 @@
 import torch
 
 from diffusers.utils import is_accelerate_available
+from packaging import version
 from transformers import CLIPFeatureExtractor, XLMRobertaTokenizer
 
 from ...configuration_utils import FrozenDict
@@ -132,6 +133,27 @@ def __init__(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
+
         self.register_modules(
             vae=vae,
             text_encoder=text_encoder,

src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion_img2img.py

@@ -20,6 +20,7 @@
 
 import PIL
 from diffusers.utils import is_accelerate_available
+from packaging import version
 from transformers import CLIPFeatureExtractor, XLMRobertaTokenizer
 
 from ...configuration_utils import FrozenDict
@@ -145,6 +146,27 @@ def __init__(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
+
         self.register_modules(
             vae=vae,
             text_encoder=text_encoder,

src/diffusers/pipelines/stable_diffusion/pipeline_cycle_diffusion.py

@@ -20,6 +20,7 @@
 
 import PIL
 from diffusers.utils import is_accelerate_available
+from packaging import version
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
 
 from ...configuration_utils import FrozenDict
@@ -176,6 +177,26 @@ def __init__(
                 "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
 
         self.register_modules(
             vae=vae,

src/diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion.py

@@ -23,6 +23,7 @@
 from flax.core.frozen_dict import FrozenDict
 from flax.jax_utils import unreplicate
 from flax.training.common_utils import shard
+from packaging import version
 from PIL import Image
 from transformers import CLIPFeatureExtractor, CLIPTokenizer, FlaxCLIPTextModel
 
@@ -34,7 +35,7 @@
     FlaxLMSDiscreteScheduler,
     FlaxPNDMScheduler,
 )
-from ...utils import logging
+from ...utils import deprecate, logging
 from . import FlaxStableDiffusionPipelineOutput
 from .safety_checker_flax import FlaxStableDiffusionSafetyChecker
 
@@ -97,6 +98,27 @@ def __init__(
                 " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
             )
 
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
+
         self.register_modules(
             vae=vae,
             text_encoder=text_encoder,

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py

@@ -18,6 +18,7 @@
 import numpy as np
 import torch
 
+from packaging import version
 from transformers import CLIPFeatureExtractor, CLIPTokenizer
 
 from ...configuration_utils import FrozenDict
@@ -98,6 +99,27 @@ def __init__(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
+
         self.register_modules(
             vae_encoder=vae_encoder,
             vae_decoder=vae_decoder,

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion_img2img.py

@@ -19,6 +19,7 @@
 import torch
 
 import PIL
+from packaging import version
 from transformers import CLIPFeatureExtractor, CLIPTokenizer
 
 from ...configuration_utils import FrozenDict
@@ -134,6 +135,27 @@ def __init__(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
+
         self.register_modules(
             vae_encoder=vae_encoder,
             vae_decoder=vae_decoder,

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion_inpaint.py

@@ -19,6 +19,7 @@
 import torch
 
 import PIL
+from packaging import version
 from transformers import CLIPFeatureExtractor, CLIPTokenizer
 
 from ...configuration_utils import FrozenDict
@@ -148,6 +149,27 @@ def __init__(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely .If you're checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
+
         self.register_modules(
             vae_encoder=vae_encoder,
             vae_decoder=vae_decoder,