[LoRA] CogView4 (#10981)

* update

* make fix-copies

* update

src/diffusers/loaders/__init__.py

@@ -70,6 +70,7 @@ if is_torch_available():
             "LoraLoaderMixin",
             "FluxLoraLoaderMixin",
             "CogVideoXLoraLoaderMixin",
+            "CogView4LoraLoaderMixin",
             "Mochi1LoraLoaderMixin",
             "HunyuanVideoLoraLoaderMixin",
             "SanaLoraLoaderMixin",
@@ -103,6 +104,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             from .lora_pipeline import (
                 AmusedLoraLoaderMixin,
                 CogVideoXLoraLoaderMixin,
+                CogView4LoraLoaderMixin,
                 FluxLoraLoaderMixin,
                 HunyuanVideoLoraLoaderMixin,
                 LoraLoaderMixin,

src/diffusers/loaders/lora_pipeline.py

@@ -4406,6 +4406,311 @@ class WanLoraLoaderMixin(LoraBaseMixin):
         super().unfuse_lora(components=components)
 
 
+class CogView4LoraLoaderMixin(LoraBaseMixin):
+    r"""
+    Load LoRA layers into [`WanTransformer3DModel`]. Specific to [`CogView4Pipeline`].
+    """
+
+    _lora_loadable_modules = ["transformer"]
+    transformer_name = TRANSFORMER_NAME
+
+    @classmethod
+    @validate_hf_hub_args
+    # Copied from diffusers.loaders.lora_pipeline.CogVideoXLoraLoaderMixin.lora_state_dict
+    def lora_state_dict(
+        cls,
+        pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]],
+        **kwargs,
+    ):
+        r"""
+        Return state dict for lora weights and the network alphas.
+
+        <Tip warning={true}>
+
+        We support loading A1111 formatted LoRA checkpoints in a limited capacity.
+
+        This function is experimental and might change in the future.
+
+        </Tip>
+
+        Parameters:
+            pretrained_model_name_or_path_or_dict (`str` or `os.PathLike` or `dict`):
+                Can be either:
+
+                    - A string, the *model id* (for example `google/ddpm-celebahq-256`) of a pretrained model hosted on
+                      the Hub.
+                    - A path to a *directory* (for example `./my_model_directory`) containing the model weights saved
+                      with [`ModelMixin.save_pretrained`].
+                    - A [torch state
+                      dict](https://pytorch.org/tutorials/beginner/saving_loading_models.html#what-is-a-state-dict).
+
+            cache_dir (`Union[str, os.PathLike]`, *optional*):
+                Path to a directory where a downloaded pretrained model configuration is cached if the standard cache
+                is not used.
+            force_download (`bool`, *optional*, defaults to `False`):
+                Whether or not to force the (re-)download of the model weights and configuration files, overriding the
+                cached versions if they exist.
+
+            proxies (`Dict[str, str]`, *optional*):
+                A dictionary of proxy servers to use by protocol or endpoint, for example, `{'http': 'foo.bar:3128',
+                'http://hostname': 'foo.bar:4012'}`. The proxies are used on each request.
+            local_files_only (`bool`, *optional*, defaults to `False`):
+                Whether to only load local model weights and configuration files or not. If set to `True`, the model
+                won't be downloaded from the Hub.
+            token (`str` or *bool*, *optional*):
+                The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
+                `diffusers-cli login` (stored in `~/.huggingface`) is used.
+            revision (`str`, *optional*, defaults to `"main"`):
+                The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
+                allowed by Git.
+            subfolder (`str`, *optional*, defaults to `""`):
+                The subfolder location of a model file within a larger model repository on the Hub or locally.
+
+        """
+        # Load the main state dict first which has the LoRA layers for either of
+        # transformer and text encoder or both.
+        cache_dir = kwargs.pop("cache_dir", None)
+        force_download = kwargs.pop("force_download", False)
+        proxies = kwargs.pop("proxies", None)
+        local_files_only = kwargs.pop("local_files_only", None)
+        token = kwargs.pop("token", None)
+        revision = kwargs.pop("revision", None)
+        subfolder = kwargs.pop("subfolder", None)
+        weight_name = kwargs.pop("weight_name", None)
+        use_safetensors = kwargs.pop("use_safetensors", None)
+
+        allow_pickle = False
+        if use_safetensors is None:
+            use_safetensors = True
+            allow_pickle = True
+
+        user_agent = {
+            "file_type": "attn_procs_weights",
+            "framework": "pytorch",
+        }
+
+        state_dict = _fetch_state_dict(
+            pretrained_model_name_or_path_or_dict=pretrained_model_name_or_path_or_dict,
+            weight_name=weight_name,
+            use_safetensors=use_safetensors,
+            local_files_only=local_files_only,
+            cache_dir=cache_dir,
+            force_download=force_download,
+            proxies=proxies,
+            token=token,
+            revision=revision,
+            subfolder=subfolder,
+            user_agent=user_agent,
+            allow_pickle=allow_pickle,
+        )
+
+        is_dora_scale_present = any("dora_scale" in k for k in state_dict)
+        if is_dora_scale_present:
+            warn_msg = "It seems like you are using a DoRA checkpoint that is not compatible in Diffusers at the moment. So, we are going to filter out the keys associated to 'dora_scale` from the state dict. If you think this is a mistake please open an issue https://github.com/huggingface/diffusers/issues/new."
+            logger.warning(warn_msg)
+            state_dict = {k: v for k, v in state_dict.items() if "dora_scale" not in k}
+
+        return state_dict
+
+    # Copied from diffusers.loaders.lora_pipeline.CogVideoXLoraLoaderMixin.load_lora_weights
+    def load_lora_weights(
+        self, pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]], adapter_name=None, **kwargs
+    ):
+        """
+        Load LoRA weights specified in `pretrained_model_name_or_path_or_dict` into `self.transformer` and
+        `self.text_encoder`. All kwargs are forwarded to `self.lora_state_dict`. See
+        [`~loaders.StableDiffusionLoraLoaderMixin.lora_state_dict`] for more details on how the state dict is loaded.
+        See [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_into_transformer`] for more details on how the state
+        dict is loaded into `self.transformer`.
+
+        Parameters:
+            pretrained_model_name_or_path_or_dict (`str` or `os.PathLike` or `dict`):
+                See [`~loaders.StableDiffusionLoraLoaderMixin.lora_state_dict`].
+            adapter_name (`str`, *optional*):
+                Adapter name to be used for referencing the loaded adapter model. If not specified, it will use
+                `default_{i}` where i is the total number of adapters being loaded.
+            low_cpu_mem_usage (`bool`, *optional*):
+                Speed up model loading by only loading the pretrained LoRA weights and not initializing the random
+                weights.
+            kwargs (`dict`, *optional*):
+                See [`~loaders.StableDiffusionLoraLoaderMixin.lora_state_dict`].
+        """
+        if not USE_PEFT_BACKEND:
+            raise ValueError("PEFT backend is required for this method.")
+
+        low_cpu_mem_usage = kwargs.pop("low_cpu_mem_usage", _LOW_CPU_MEM_USAGE_DEFAULT_LORA)
+        if low_cpu_mem_usage and is_peft_version("<", "0.13.0"):
+            raise ValueError(
+                "`low_cpu_mem_usage=True` is not compatible with this `peft` version. Please update it with `pip install -U peft`."
+            )
+
+        # if a dict is passed, copy it instead of modifying it inplace
+        if isinstance(pretrained_model_name_or_path_or_dict, dict):
+            pretrained_model_name_or_path_or_dict = pretrained_model_name_or_path_or_dict.copy()
+
+        # First, ensure that the checkpoint is a compatible one and can be successfully loaded.
+        state_dict = self.lora_state_dict(pretrained_model_name_or_path_or_dict, **kwargs)
+
+        is_correct_format = all("lora" in key for key in state_dict.keys())
+        if not is_correct_format:
+            raise ValueError("Invalid LoRA checkpoint.")
+
+        self.load_lora_into_transformer(
+            state_dict,
+            transformer=getattr(self, self.transformer_name) if not hasattr(self, "transformer") else self.transformer,
+            adapter_name=adapter_name,
+            _pipeline=self,
+            low_cpu_mem_usage=low_cpu_mem_usage,
+        )
+
+    @classmethod
+    # Copied from diffusers.loaders.lora_pipeline.SD3LoraLoaderMixin.load_lora_into_transformer with SD3Transformer2DModel->CogView4Transformer2DModel
+    def load_lora_into_transformer(
+        cls, state_dict, transformer, adapter_name=None, _pipeline=None, low_cpu_mem_usage=False
+    ):
+        """
+        This will load the LoRA layers specified in `state_dict` into `transformer`.
+
+        Parameters:
+            state_dict (`dict`):
+                A standard state dict containing the lora layer parameters. The keys can either be indexed directly
+                into the unet or prefixed with an additional `unet` which can be used to distinguish between text
+                encoder lora layers.
+            transformer (`CogView4Transformer2DModel`):
+                The Transformer model to load the LoRA layers into.
+            adapter_name (`str`, *optional*):
+                Adapter name to be used for referencing the loaded adapter model. If not specified, it will use
+                `default_{i}` where i is the total number of adapters being loaded.
+            low_cpu_mem_usage (`bool`, *optional*):
+                Speed up model loading by only loading the pretrained LoRA weights and not initializing the random
+                weights.
+        """
+        if low_cpu_mem_usage and is_peft_version("<", "0.13.0"):
+            raise ValueError(
+                "`low_cpu_mem_usage=True` is not compatible with this `peft` version. Please update it with `pip install -U peft`."
+            )
+
+        # Load the layers corresponding to transformer.
+        logger.info(f"Loading {cls.transformer_name}.")
+        transformer.load_lora_adapter(
+            state_dict,
+            network_alphas=None,
+            adapter_name=adapter_name,
+            _pipeline=_pipeline,
+            low_cpu_mem_usage=low_cpu_mem_usage,
+        )
+
+    @classmethod
+    # Copied from diffusers.loaders.lora_pipeline.CogVideoXLoraLoaderMixin.save_lora_weights
+    def save_lora_weights(
+        cls,
+        save_directory: Union[str, os.PathLike],
+        transformer_lora_layers: Dict[str, Union[torch.nn.Module, torch.Tensor]] = None,
+        is_main_process: bool = True,
+        weight_name: str = None,
+        save_function: Callable = None,
+        safe_serialization: bool = True,
+    ):
+        r"""
+        Save the LoRA parameters corresponding to the UNet and text encoder.
+
+        Arguments:
+            save_directory (`str` or `os.PathLike`):
+                Directory to save LoRA parameters to. Will be created if it doesn't exist.
+            transformer_lora_layers (`Dict[str, torch.nn.Module]` or `Dict[str, torch.Tensor]`):
+                State dict of the LoRA layers corresponding to the `transformer`.
+            is_main_process (`bool`, *optional*, defaults to `True`):
+                Whether the process calling this is the main process or not. Useful during distributed training and you
+                need to call this function on all processes. In this case, set `is_main_process=True` only on the main
+                process to avoid race conditions.
+            save_function (`Callable`):
+                The function to use to save the state dictionary. Useful during distributed training when you need to
+                replace `torch.save` with another method. Can be configured with the environment variable
+                `DIFFUSERS_SAVE_MODE`.
+            safe_serialization (`bool`, *optional*, defaults to `True`):
+                Whether to save the model using `safetensors` or the traditional PyTorch way with `pickle`.
+        """
+        state_dict = {}
+
+        if not transformer_lora_layers:
+            raise ValueError("You must pass `transformer_lora_layers`.")
+
+        if transformer_lora_layers:
+            state_dict.update(cls.pack_weights(transformer_lora_layers, cls.transformer_name))
+
+        # Save the model
+        cls.write_lora_layers(
+            state_dict=state_dict,
+            save_directory=save_directory,
+            is_main_process=is_main_process,
+            weight_name=weight_name,
+            save_function=save_function,
+            safe_serialization=safe_serialization,
+        )
+
+    # Copied from diffusers.loaders.lora_pipeline.CogVideoXLoraLoaderMixin.fuse_lora
+    def fuse_lora(
+        self,
+        components: List[str] = ["transformer"],
+        lora_scale: float = 1.0,
+        safe_fusing: bool = False,
+        adapter_names: Optional[List[str]] = None,
+        **kwargs,
+    ):
+        r"""
+        Fuses the LoRA parameters into the original parameters of the corresponding blocks.
+
+        <Tip warning={true}>
+
+        This is an experimental API.
+
+        </Tip>
+
+        Args:
+            components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
+            lora_scale (`float`, defaults to 1.0):
+                Controls how much to influence the outputs with the LoRA parameters.
+            safe_fusing (`bool`, defaults to `False`):
+                Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
+            adapter_names (`List[str]`, *optional*):
+                Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
+
+        Example:
+
+        ```py
+        from diffusers import DiffusionPipeline
+        import torch
+
+        pipeline = DiffusionPipeline.from_pretrained(
+            "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
+        ).to("cuda")
+        pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
+        pipeline.fuse_lora(lora_scale=0.7)
+        ```
+        """
+        super().fuse_lora(
+            components=components, lora_scale=lora_scale, safe_fusing=safe_fusing, adapter_names=adapter_names
+        )
+
+    # Copied from diffusers.loaders.lora_pipeline.CogVideoXLoraLoaderMixin.unfuse_lora
+    def unfuse_lora(self, components: List[str] = ["transformer"], **kwargs):
+        r"""
+        Reverses the effect of
+        [`pipe.fuse_lora()`](https://huggingface.co/docs/diffusers/main/en/api/loaders#diffusers.loaders.LoraBaseMixin.fuse_lora).
+
+        <Tip warning={true}>
+
+        This is an experimental API.
+
+        </Tip>
+
+        Args:
+            components (`List[str]`): List of LoRA-injectable components to unfuse LoRA from.
+            unfuse_transformer (`bool`, defaults to `True`): Whether to unfuse the UNet LoRA parameters.
+        """
+        super().unfuse_lora(components=components)
+
+
 class LoraLoaderMixin(StableDiffusionLoraLoaderMixin):
     def __init__(self, *args, **kwargs):
         deprecation_message = "LoraLoaderMixin is deprecated and this will be removed in a future version. Please use `StableDiffusionLoraLoaderMixin`, instead."

src/diffusers/loaders/peft.py

@@ -54,6 +54,7 @@ _SET_ADAPTER_SCALE_FN_MAPPING = {
     "SanaTransformer2DModel": lambda model_cls, weights: weights,
     "Lumina2Transformer2DModel": lambda model_cls, weights: weights,
     "WanTransformer3DModel": lambda model_cls, weights: weights,
+    "CogView4Transformer2DModel": lambda model_cls, weights: weights,
 }
 
 

src/diffusers/models/transformers/transformer_cogview4.py

@@ -12,20 +12,21 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Optional, Tuple, Union
+from typing import Any, Dict, Optional, Tuple, Union
 
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
 from ...configuration_utils import ConfigMixin, register_to_config
-from ...models.attention import FeedForward
-from ...models.attention_processor import Attention
-from ...models.modeling_utils import ModelMixin
-from ...models.normalization import AdaLayerNormContinuous
-from ...utils import logging
+from ...loaders import PeftAdapterMixin
+from ...utils import USE_PEFT_BACKEND, logging, scale_lora_layers, unscale_lora_layers
+from ..attention import FeedForward
+from ..attention_processor import Attention
 from ..embeddings import CogView3CombinedTimestepSizeEmbeddings
 from ..modeling_outputs import Transformer2DModelOutput
+from ..modeling_utils import ModelMixin
+from ..normalization import AdaLayerNormContinuous
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -288,7 +289,7 @@ class CogView4RotaryPosEmbed(nn.Module):
         return (freqs.cos(), freqs.sin())
 
 
-class CogView4Transformer2DModel(ModelMixin, ConfigMixin):
+class CogView4Transformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
     r"""
     Args:
         patch_size (`int`, defaults to `2`):
@@ -383,8 +384,24 @@ class CogView4Transformer2DModel(ModelMixin, ConfigMixin):
         original_size: torch.Tensor,
         target_size: torch.Tensor,
         crop_coords: torch.Tensor,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
         return_dict: bool = True,
     ) -> Union[torch.Tensor, Transformer2DModelOutput]:
+        if attention_kwargs is not None:
+            attention_kwargs = attention_kwargs.copy()
+            lora_scale = attention_kwargs.pop("scale", 1.0)
+        else:
+            lora_scale = 1.0
+
+        if USE_PEFT_BACKEND:
+            # weight the lora layers by setting `lora_scale` for each PEFT layer
+            scale_lora_layers(self, lora_scale)
+        else:
+            if attention_kwargs is not None and attention_kwargs.get("scale", None) is not None:
+                logger.warning(
+                    "Passing `scale` via `attention_kwargs` when not using the PEFT backend is ineffective."
+                )
+
         batch_size, num_channels, height, width = hidden_states.shape
 
         # 1. RoPE
@@ -419,6 +436,10 @@ class CogView4Transformer2DModel(ModelMixin, ConfigMixin):
         hidden_states = hidden_states.reshape(batch_size, post_patch_height, post_patch_width, -1, p, p)
         output = hidden_states.permute(0, 3, 1, 4, 2, 5).flatten(4, 5).flatten(2, 3)
 
+        if USE_PEFT_BACKEND:
+            # remove `lora_scale` from each PEFT layer
+            unscale_lora_layers(self, lora_scale)
+
         if not return_dict:
             return (output,)
         return Transformer2DModelOutput(sample=output)

src/diffusers/pipelines/cogview4/pipeline_cogview4.py

@@ -14,7 +14,7 @@
 # limitations under the License.
 
 import inspect
-from typing import Callable, Dict, List, Optional, Tuple, Union
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 
 import numpy as np
 import torch
@@ -22,6 +22,7 @@ from transformers import AutoTokenizer, GlmModel
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import VaeImageProcessor
+from ...loaders import CogView4LoraLoaderMixin
 from ...models import AutoencoderKL, CogView4Transformer2DModel
 from ...pipelines.pipeline_utils import DiffusionPipeline
 from ...schedulers import FlowMatchEulerDiscreteScheduler
@@ -133,7 +134,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class CogView4Pipeline(DiffusionPipeline):
+class CogView4Pipeline(DiffusionPipeline, CogView4LoraLoaderMixin):
     r"""
     Pipeline for text-to-image generation using CogView4.
 
@@ -392,6 +393,10 @@ class CogView4Pipeline(DiffusionPipeline):
     def interrupt(self):
         return self._interrupt
 
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
@@ -413,6 +418,7 @@ class CogView4Pipeline(DiffusionPipeline):
         crops_coords_top_left: Tuple[int, int] = (0, 0),
         output_type: str = "pil",
         return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
         callback_on_step_end: Optional[
             Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
         ] = None,
@@ -526,6 +532,7 @@ class CogView4Pipeline(DiffusionPipeline):
             negative_prompt_embeds,
         )
         self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
         self._interrupt = False
 
         # Default call parameters
@@ -615,6 +622,7 @@ class CogView4Pipeline(DiffusionPipeline):
                     original_size=original_size,
                     target_size=target_size,
                     crop_coords=crops_coords_top_left,
+                    attention_kwargs=attention_kwargs,
                     return_dict=False,
                 )[0]
 
@@ -627,6 +635,7 @@ class CogView4Pipeline(DiffusionPipeline):
                         original_size=original_size,
                         target_size=target_size,
                         crop_coords=crops_coords_top_left,
+                        attention_kwargs=attention_kwargs,
                         return_dict=False,
                     )[0]
 

tests/lora/test_lora_layers_cogview4.py

+# Copyright 2024 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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 sys
+import tempfile
+import unittest
+
+import numpy as np
+import torch
+from transformers import AutoTokenizer, GlmModel
+
+from diffusers import AutoencoderKL, CogView4Pipeline, CogView4Transformer2DModel, FlowMatchEulerDiscreteScheduler
+from diffusers.utils.testing_utils import floats_tensor, require_peft_backend, skip_mps, torch_device
+
+
+sys.path.append(".")
+
+from utils import PeftLoraLoaderMixinTests  # noqa: E402
+
+
+class TokenizerWrapper:
+    @staticmethod
+    def from_pretrained(*args, **kwargs):
+        return AutoTokenizer.from_pretrained(
+            "hf-internal-testing/tiny-random-cogview4", subfolder="tokenizer", trust_remote_code=True
+        )
+
+
+@require_peft_backend
+@skip_mps
+class CogView4LoRATests(unittest.TestCase, PeftLoraLoaderMixinTests):
+    pipeline_class = CogView4Pipeline
+    scheduler_cls = FlowMatchEulerDiscreteScheduler
+    scheduler_classes = [FlowMatchEulerDiscreteScheduler]
+    scheduler_kwargs = {}
+
+    transformer_kwargs = {
+        "patch_size": 2,
+        "in_channels": 4,
+        "num_layers": 2,
+        "attention_head_dim": 4,
+        "num_attention_heads": 4,
+        "out_channels": 4,
+        "text_embed_dim": 32,
+        "time_embed_dim": 8,
+        "condition_dim": 4,
+    }
+    transformer_cls = CogView4Transformer2DModel
+    vae_kwargs = {
+        "block_out_channels": [32, 64],
+        "in_channels": 3,
+        "out_channels": 3,
+        "down_block_types": ["DownEncoderBlock2D", "DownEncoderBlock2D"],
+        "up_block_types": ["UpDecoderBlock2D", "UpDecoderBlock2D"],
+        "latent_channels": 4,
+        "sample_size": 128,
+    }
+    vae_cls = AutoencoderKL
+    tokenizer_cls, tokenizer_id, tokenizer_subfolder = (
+        TokenizerWrapper,
+        "hf-internal-testing/tiny-random-cogview4",
+        "tokenizer",
+    )
+    text_encoder_cls, text_encoder_id, text_encoder_subfolder = (
+        GlmModel,
+        "hf-internal-testing/tiny-random-cogview4",
+        "text_encoder",
+    )
+
+    @property
+    def output_shape(self):
+        return (1, 32, 32, 3)
+
+    def get_dummy_inputs(self, with_generator=True):
+        batch_size = 1
+        sequence_length = 16
+        num_channels = 4
+        sizes = (4, 4)
+
+        generator = torch.manual_seed(0)
+        noise = floats_tensor((batch_size, num_channels) + sizes)
+        input_ids = torch.randint(1, sequence_length, size=(batch_size, sequence_length), generator=generator)
+
+        pipeline_inputs = {
+            "prompt": "",
+            "num_inference_steps": 1,
+            "guidance_scale": 6.0,
+            "height": 32,
+            "width": 32,
+            "max_sequence_length": sequence_length,
+            "output_type": "np",
+        }
+        if with_generator:
+            pipeline_inputs.update({"generator": generator})
+
+        return noise, input_ids, pipeline_inputs
+
+    def test_simple_inference_with_text_lora_denoiser_fused_multi(self):
+        super().test_simple_inference_with_text_lora_denoiser_fused_multi(expected_atol=9e-3)
+
+    def test_simple_inference_with_text_denoiser_lora_unfused(self):
+        super().test_simple_inference_with_text_denoiser_lora_unfused(expected_atol=9e-3)
+
+    def test_simple_inference_save_pretrained(self):
+        """
+        Tests a simple usecase where users could use saving utilities for LoRA through save_pretrained
+        """
+        for scheduler_cls in self.scheduler_classes:
+            components, _, _ = self.get_dummy_components(scheduler_cls)
+            pipe = self.pipeline_class(**components)
+            pipe = pipe.to(torch_device)
+            pipe.set_progress_bar_config(disable=None)
+            _, _, inputs = self.get_dummy_inputs(with_generator=False)
+
+            output_no_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
+            self.assertTrue(output_no_lora.shape == self.output_shape)
+
+            images_lora = pipe(**inputs, generator=torch.manual_seed(0))[0]
+
+            with tempfile.TemporaryDirectory() as tmpdirname:
+                pipe.save_pretrained(tmpdirname)
+
+                pipe_from_pretrained = self.pipeline_class.from_pretrained(tmpdirname)
+                pipe_from_pretrained.to(torch_device)
+
+            images_lora_save_pretrained = pipe_from_pretrained(**inputs, generator=torch.manual_seed(0))[0]
+
+            self.assertTrue(
+                np.allclose(images_lora, images_lora_save_pretrained, atol=1e-3, rtol=1e-3),
+                "Loading from saved checkpoints should give same results.",
+            )
+
+    @unittest.skip("Not supported in CogView4.")
+    def test_simple_inference_with_text_denoiser_block_scale(self):
+        pass
+
+    @unittest.skip("Not supported in CogView4.")
+    def test_simple_inference_with_text_denoiser_block_scale_for_all_dict_options(self):
+        pass
+
+    @unittest.skip("Not supported in CogView4.")
+    def test_modify_padding_mode(self):
+        pass
+
+    @unittest.skip("Text encoder LoRA is not supported in CogView4.")
+    def test_simple_inference_with_partial_text_lora(self):
+        pass
+
+    @unittest.skip("Text encoder LoRA is not supported in CogView4.")
+    def test_simple_inference_with_text_lora(self):
+        pass
+
+    @unittest.skip("Text encoder LoRA is not supported in CogView4.")
+    def test_simple_inference_with_text_lora_and_scale(self):
+        pass
+
+    @unittest.skip("Text encoder LoRA is not supported in CogView4.")
+    def test_simple_inference_with_text_lora_fused(self):
+        pass
+
+    @unittest.skip("Text encoder LoRA is not supported in CogView4.")
+    def test_simple_inference_with_text_lora_save_load(self):
+        pass