diffusers 0.30.2__py3-none-any.whl → 0.31.0__py3-none-any.whl

diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_inpaint.py

@@ -25,7 +25,7 @@ from transformers import (
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import SD3LoraLoaderMixin
+from ...loaders import FromSingleFileMixin, SD3LoraLoaderMixin
 from ...models.autoencoders import AutoencoderKL
 from ...models.transformers import SD3Transformer2DModel
 from ...schedulers import FlowMatchEulerDiscreteScheduler
@@ -97,7 +97,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -148,7 +148,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class StableDiffusion3InpaintPipeline(DiffusionPipeline):
+class StableDiffusion3InpaintPipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingleFileMixin):
     r"""
     Args:
         transformer ([`SD3Transformer2DModel`]):

diffusers/pipelines/stable_diffusion_k_diffusion/pipeline_stable_diffusion_k_diffusion.py

@@ -602,9 +602,9 @@ class StableDiffusionKDiffusionPipeline(
             sigma_min: float = self.k_diffusion_model.sigmas[0].item()
             sigma_max: float = self.k_diffusion_model.sigmas[-1].item()
             sigmas = get_sigmas_karras(n=num_inference_steps, sigma_min=sigma_min, sigma_max=sigma_max)
-            sigmas = sigmas.to(device)
         else:
             sigmas = self.scheduler.sigmas
+        sigmas = sigmas.to(device)
         sigmas = sigmas.to(prompt_embeds.dtype)
 
         # 6. Prepare latent variables

diffusers/pipelines/stable_diffusion_ldm3d/pipeline_stable_diffusion_ldm3d.py

@@ -61,9 +61,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -83,7 +95,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py

@@ -61,9 +61,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -83,7 +95,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py

@@ -87,9 +87,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -109,7 +121,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py

@@ -90,9 +90,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -126,7 +138,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -640,14 +652,16 @@ class StableDiffusionXLImg2ImgPipeline(
         if denoising_start is None:
             init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
             t_start = max(num_inference_steps - init_timestep, 0)
-        else:
-            t_start = 0
 
-        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start * self.scheduler.order)
 
-        # Strength is irrelevant if we directly request a timestep to start at;
-        # that is, strength is determined by the denoising_start instead.
-        if denoising_start is not None:
+            return timesteps, num_inference_steps - t_start
+
+        else:
+            # Strength is irrelevant if we directly request a timestep to start at;
+            # that is, strength is determined by the denoising_start instead.
             discrete_timestep_cutoff = int(
                 round(
                     self.scheduler.config.num_train_timesteps
@@ -655,7 +669,7 @@ class StableDiffusionXLImg2ImgPipeline(
                 )
             )
 
-            num_inference_steps = (timesteps < discrete_timestep_cutoff).sum().item()
+            num_inference_steps = (self.scheduler.timesteps < discrete_timestep_cutoff).sum().item()
             if self.scheduler.order == 2 and num_inference_steps % 2 == 0:
                 # if the scheduler is a 2nd order scheduler we might have to do +1
                 # because `num_inference_steps` might be even given that every timestep
@@ -666,11 +680,12 @@ class StableDiffusionXLImg2ImgPipeline(
                 num_inference_steps = num_inference_steps + 1
 
             # because t_n+1 >= t_n, we slice the timesteps starting from the end
-            timesteps = timesteps[-num_inference_steps:]
+            t_start = len(self.scheduler.timesteps) - num_inference_steps
+            timesteps = self.scheduler.timesteps[t_start:]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start)
             return timesteps, num_inference_steps
 
-        return timesteps, num_inference_steps - t_start
-
     def prepare_latents(
         self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True
     ):

diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py

@@ -101,9 +101,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -153,7 +165,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -901,14 +913,16 @@ class StableDiffusionXLInpaintPipeline(
         if denoising_start is None:
             init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
             t_start = max(num_inference_steps - init_timestep, 0)
-        else:
-            t_start = 0
 
-        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start * self.scheduler.order)
 
-        # Strength is irrelevant if we directly request a timestep to start at;
-        # that is, strength is determined by the denoising_start instead.
-        if denoising_start is not None:
+            return timesteps, num_inference_steps - t_start
+
+        else:
+            # Strength is irrelevant if we directly request a timestep to start at;
+            # that is, strength is determined by the denoising_start instead.
             discrete_timestep_cutoff = int(
                 round(
                     self.scheduler.config.num_train_timesteps
@@ -916,7 +930,7 @@ class StableDiffusionXLInpaintPipeline(
                 )
             )
 
-            num_inference_steps = (timesteps < discrete_timestep_cutoff).sum().item()
+            num_inference_steps = (self.scheduler.timesteps < discrete_timestep_cutoff).sum().item()
             if self.scheduler.order == 2 and num_inference_steps % 2 == 0:
                 # if the scheduler is a 2nd order scheduler we might have to do +1
                 # because `num_inference_steps` might be even given that every timestep
@@ -927,11 +941,12 @@ class StableDiffusionXLInpaintPipeline(
                 num_inference_steps = num_inference_steps + 1
 
             # because t_n+1 >= t_n, we slice the timesteps starting from the end
-            timesteps = timesteps[-num_inference_steps:]
+            t_start = len(self.scheduler.timesteps) - num_inference_steps
+            timesteps = self.scheduler.timesteps[t_start:]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start)
             return timesteps, num_inference_steps
 
-        return timesteps, num_inference_steps - t_start
-
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids
     def _get_add_time_ids(
         self,

diffusers/pipelines/stable_video_diffusion/pipeline_stable_video_diffusion.py

@@ -71,7 +71,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py

@@ -127,7 +127,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py

@@ -119,9 +119,21 @@ def _preprocess_adapter_image(image, height, width):
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -141,7 +153,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero_sdxl.py

@@ -310,9 +310,21 @@ def create_motion_field_and_warp_latents(motion_field_strength_x, motion_field_s
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)

diffusers/quantizers/__init__.py

@@ -0,0 +1,16 @@
+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# 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.
+
+from .auto import DiffusersAutoQuantizer
+from .base import DiffusersQuantizer

diffusers/quantizers/auto.py

@@ -0,0 +1,126 @@
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# 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.
+"""
+Adapted from
+https://github.com/huggingface/transformers/blob/c409cd81777fb27aadc043ed3d8339dbc020fb3b/src/transformers/quantizers/auto.py
+"""
+import warnings
+from typing import Dict, Optional, Union
+
+from .bitsandbytes import BnB4BitDiffusersQuantizer, BnB8BitDiffusersQuantizer
+from .quantization_config import BitsAndBytesConfig, QuantizationConfigMixin, QuantizationMethod
+
+
+AUTO_QUANTIZER_MAPPING = {
+    "bitsandbytes_4bit": BnB4BitDiffusersQuantizer,
+    "bitsandbytes_8bit": BnB8BitDiffusersQuantizer,
+}
+
+AUTO_QUANTIZATION_CONFIG_MAPPING = {
+    "bitsandbytes_4bit": BitsAndBytesConfig,
+    "bitsandbytes_8bit": BitsAndBytesConfig,
+}
+
+
+class DiffusersAutoQuantizer:
+    """
+     The auto diffusers quantizer class that takes care of automatically instantiating to the correct
+    `DiffusersQuantizer` given the `QuantizationConfig`.
+    """
+
+    @classmethod
+    def from_dict(cls, quantization_config_dict: Dict):
+        quant_method = quantization_config_dict.get("quant_method", None)
+        # We need a special care for bnb models to make sure everything is BC ..
+        if quantization_config_dict.get("load_in_8bit", False) or quantization_config_dict.get("load_in_4bit", False):
+            suffix = "_4bit" if quantization_config_dict.get("load_in_4bit", False) else "_8bit"
+            quant_method = QuantizationMethod.BITS_AND_BYTES + suffix
+        elif quant_method is None:
+            raise ValueError(
+                "The model's quantization config from the arguments has no `quant_method` attribute. Make sure that the model has been correctly quantized"
+            )
+
+        if quant_method not in AUTO_QUANTIZATION_CONFIG_MAPPING.keys():
+            raise ValueError(
+                f"Unknown quantization type, got {quant_method} - supported types are:"
+                f" {list(AUTO_QUANTIZER_MAPPING.keys())}"
+            )
+
+        target_cls = AUTO_QUANTIZATION_CONFIG_MAPPING[quant_method]
+        return target_cls.from_dict(quantization_config_dict)
+
+    @classmethod
+    def from_config(cls, quantization_config: Union[QuantizationConfigMixin, Dict], **kwargs):
+        # Convert it to a QuantizationConfig if the q_config is a dict
+        if isinstance(quantization_config, dict):
+            quantization_config = cls.from_dict(quantization_config)
+
+        quant_method = quantization_config.quant_method
+
+        # Again, we need a special care for bnb as we have a single quantization config
+        # class for both 4-bit and 8-bit quantization
+        if quant_method == QuantizationMethod.BITS_AND_BYTES:
+            if quantization_config.load_in_8bit:
+                quant_method += "_8bit"
+            else:
+                quant_method += "_4bit"
+
+        if quant_method not in AUTO_QUANTIZER_MAPPING.keys():
+            raise ValueError(
+                f"Unknown quantization type, got {quant_method} - supported types are:"
+                f" {list(AUTO_QUANTIZER_MAPPING.keys())}"
+            )
+
+        target_cls = AUTO_QUANTIZER_MAPPING[quant_method]
+        return target_cls(quantization_config, **kwargs)
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
+        model_config = cls.load_config(pretrained_model_name_or_path, **kwargs)
+        if getattr(model_config, "quantization_config", None) is None:
+            raise ValueError(
+                f"Did not found a `quantization_config` in {pretrained_model_name_or_path}. Make sure that the model is correctly quantized."
+            )
+        quantization_config_dict = model_config.quantization_config
+        quantization_config = cls.from_dict(quantization_config_dict)
+        # Update with potential kwargs that are passed through from_pretrained.
+        quantization_config.update(kwargs)
+
+        return cls.from_config(quantization_config)
+
+    @classmethod
+    def merge_quantization_configs(
+        cls,
+        quantization_config: Union[dict, QuantizationConfigMixin],
+        quantization_config_from_args: Optional[QuantizationConfigMixin],
+    ):
+        """
+        handles situations where both quantization_config from args and quantization_config from model config are
+        present.
+        """
+        if quantization_config_from_args is not None:
+            warning_msg = (
+                "You passed `quantization_config` or equivalent parameters to `from_pretrained` but the model you're loading"
+                " already has a `quantization_config` attribute. The `quantization_config` from the model will be used."
+            )
+        else:
+            warning_msg = ""
+
+        if isinstance(quantization_config, dict):
+            quantization_config = cls.from_dict(quantization_config)
+
+        if warning_msg != "":
+            warnings.warn(warning_msg)
+
+        return quantization_config