optimum-rbln 0.9.3.post1__py3-none-any.whl

optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_3.py

@@ -0,0 +1,176 @@
+# Copyright 2025 Rebellions Inc. 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 typing import Any, Optional, Tuple
+
+from ....configuration_utils import RBLNModelConfig
+from ....transformers import RBLNCLIPTextModelWithProjectionConfig, RBLNT5EncoderModelConfig
+from ..models import RBLNAutoencoderKLConfig, RBLNSD3Transformer2DModelConfig
+
+
+class RBLNStableDiffusion3PipelineBaseConfig(RBLNModelConfig):
+    submodules = ["transformer", "text_encoder", "text_encoder_2", "text_encoder_3", "vae"]
+    _vae_uses_encoder = False
+
+    def __init__(
+        self,
+        transformer: Optional[RBLNSD3Transformer2DModelConfig] = None,
+        text_encoder: Optional[RBLNCLIPTextModelWithProjectionConfig] = None,
+        text_encoder_2: Optional[RBLNCLIPTextModelWithProjectionConfig] = None,
+        text_encoder_3: Optional[RBLNT5EncoderModelConfig] = None,
+        vae: Optional[RBLNAutoencoderKLConfig] = None,
+        *,
+        max_seq_len: Optional[int] = None,
+        sample_size: Optional[Tuple[int, int]] = None,
+        image_size: Optional[Tuple[int, int]] = None,
+        batch_size: Optional[int] = None,
+        img_height: Optional[int] = None,
+        img_width: Optional[int] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        guidance_scale: Optional[float] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            transformer (Optional[RBLNSD3Transformer2DModelConfig]): Configuration for the transformer model component.
+                Initialized as RBLNSD3Transformer2DModelConfig if not provided.
+            text_encoder (Optional[RBLNCLIPTextModelWithProjectionConfig]): Configuration for the primary text encoder.
+                Initialized as RBLNCLIPTextModelWithProjectionConfig if not provided.
+            text_encoder_2 (Optional[RBLNCLIPTextModelWithProjectionConfig]): Configuration for the secondary text encoder.
+                Initialized as RBLNCLIPTextModelWithProjectionConfig if not provided.
+            text_encoder_3 (Optional[RBLNT5EncoderModelConfig]): Configuration for the tertiary text encoder.
+                Initialized as RBLNT5EncoderModelConfig if not provided.
+            vae (Optional[RBLNAutoencoderKLConfig]): Configuration for the VAE model component.
+                Initialized as RBLNAutoencoderKLConfig if not provided.
+            max_seq_len (Optional[int]): Maximum sequence length for text inputs. Defaults to 256.
+            sample_size (Optional[Tuple[int, int]]): Spatial dimensions for the transformer model.
+            image_size (Optional[Tuple[int, int]]): Dimensions for the generated images.
+                Cannot be used together with img_height/img_width.
+            batch_size (Optional[int]): Batch size for inference, applied to all submodules.
+            img_height (Optional[int]): Height of the generated images.
+            img_width (Optional[int]): Width of the generated images.
+            height (Optional[int]): Height of the generated images.
+            width (Optional[int]): Width of the generated images.
+            guidance_scale (Optional[float]): Scale for classifier-free guidance.
+            kwargs: Additional arguments passed to the parent RBLNModelConfig.
+
+        Raises:
+            ValueError: If both image_size and img_height/img_width are provided.
+
+        Note:
+            When guidance_scale > 1.0, the transformer batch size is automatically doubled to
+            accommodate classifier-free guidance.
+        """
+        super().__init__(**kwargs)
+
+        # Initial check for image_size conflict remains as is
+        if image_size is not None and (
+            img_height is not None or img_width is not None or height is not None or width is not None
+        ):
+            raise ValueError("image_size cannot be provided alongside img_height/img_width or height/width")
+
+        # Prioritize height/width (HF-aligned)
+        if height is not None and width is not None:
+            if img_height is not None or img_width is not None:
+                # Raise error if both sets of arguments are provided
+                raise ValueError(
+                    "Cannot provide both 'height'/'width' and 'img_height'/'img_width' simultaneously. "
+                    "Please use one set of arguments for image dimensions, preferring 'height'/'width'."
+                )
+            image_size = (height, width)
+        elif (height is not None and width is None) or (height is None and width is not None):
+            raise ValueError("Both height and width must be provided together if used")
+        # Fallback to img_height/img_width for backward compatibility
+        elif img_height is not None and img_width is not None:
+            image_size = (img_height, img_width)
+        elif (img_height is not None and img_width is None) or (img_height is None and img_width is not None):
+            raise ValueError("Both img_height and img_width must be provided together if used")
+
+        max_seq_len = max_seq_len or 256
+
+        self.text_encoder = self.initialize_submodule_config(
+            text_encoder,
+            cls_name="RBLNCLIPTextModelWithProjectionConfig",
+            batch_size=batch_size,
+        )
+        self.text_encoder_2 = self.initialize_submodule_config(
+            text_encoder_2,
+            cls_name="RBLNCLIPTextModelWithProjectionConfig",
+            batch_size=batch_size,
+        )
+        self.text_encoder_3 = self.initialize_submodule_config(
+            text_encoder_3,
+            cls_name="RBLNT5EncoderModelConfig",
+            batch_size=batch_size,
+            max_seq_len=max_seq_len,
+            model_input_names=["input_ids"],
+        )
+        self.transformer = self.initialize_submodule_config(
+            transformer,
+            cls_name="RBLNSD3Transformer2DModelConfig",
+            sample_size=sample_size,
+        )
+        self.vae = self.initialize_submodule_config(
+            vae,
+            cls_name="RBLNAutoencoderKLConfig",
+            batch_size=batch_size,
+            uses_encoder=self.__class__._vae_uses_encoder,
+            sample_size=image_size,
+        )
+
+        # Get default guidance scale from original class to set Transformer batch size
+        if guidance_scale is None:
+            guidance_scale = self.get_default_values_for_original_cls("__call__", ["guidance_scale"])["guidance_scale"]
+
+        if not self.transformer.batch_size_is_specified:
+            do_classifier_free_guidance = guidance_scale > 1.0
+            if do_classifier_free_guidance:
+                self.transformer.batch_size = self.text_encoder.batch_size * 2
+            else:
+                self.transformer.batch_size = self.text_encoder.batch_size
+
+    @property
+    def max_seq_len(self):
+        return self.text_encoder_3.max_seq_len
+
+    @property
+    def batch_size(self):
+        return self.vae.batch_size
+
+    @property
+    def sample_size(self):
+        return self.transformer.sample_size
+
+    @property
+    def image_size(self):
+        return self.vae.sample_size
+
+
+class RBLNStableDiffusion3PipelineConfig(RBLNStableDiffusion3PipelineBaseConfig):
+    """Config for SD3 Text2Img Pipeline"""
+
+    _vae_uses_encoder = False
+
+
+class RBLNStableDiffusion3Img2ImgPipelineConfig(RBLNStableDiffusion3PipelineBaseConfig):
+    """Config for SD3 Img2Img Pipeline"""
+
+    _vae_uses_encoder = True
+
+
+class RBLNStableDiffusion3InpaintPipelineConfig(RBLNStableDiffusion3PipelineBaseConfig):
+    """Config for SD3 Inpainting Pipeline"""
+
+    _vae_uses_encoder = True

optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_xl.py

@@ -0,0 +1,159 @@
+# Copyright 2025 Rebellions Inc. 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 typing import Any, Optional, Tuple
+
+from ....configuration_utils import RBLNModelConfig
+from ....transformers import RBLNCLIPTextModelConfig, RBLNCLIPTextModelWithProjectionConfig
+from ..models import RBLNAutoencoderKLConfig, RBLNUNet2DConditionModelConfig
+
+
+class RBLNStableDiffusionXLPipelineBaseConfig(RBLNModelConfig):
+    submodules = ["text_encoder", "text_encoder_2", "unet", "vae"]
+    _vae_uses_encoder = False
+
+    def __init__(
+        self,
+        text_encoder: Optional[RBLNCLIPTextModelConfig] = None,
+        text_encoder_2: Optional[RBLNCLIPTextModelWithProjectionConfig] = None,
+        unet: Optional[RBLNUNet2DConditionModelConfig] = None,
+        vae: Optional[RBLNAutoencoderKLConfig] = None,
+        *,
+        batch_size: Optional[int] = None,
+        img_height: Optional[int] = None,
+        img_width: Optional[int] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        sample_size: Optional[Tuple[int, int]] = None,
+        image_size: Optional[Tuple[int, int]] = None,
+        guidance_scale: Optional[float] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            text_encoder (Optional[RBLNCLIPTextModelConfig]): Configuration for the primary text encoder component.
+                Initialized as RBLNCLIPTextModelConfig if not provided.
+            text_encoder_2 (Optional[RBLNCLIPTextModelWithProjectionConfig]): Configuration for the secondary text encoder component.
+                Initialized as RBLNCLIPTextModelWithProjectionConfig if not provided.
+            unet (Optional[RBLNUNet2DConditionModelConfig]): Configuration for the UNet model component.
+                Initialized as RBLNUNet2DConditionModelConfig if not provided.
+            vae (Optional[RBLNAutoencoderKLConfig]): Configuration for the VAE model component.
+                Initialized as RBLNAutoencoderKLConfig if not provided.
+            batch_size (Optional[int]): Batch size for inference, applied to all submodules.
+            img_height (Optional[int]): Height of the generated images.
+            img_width (Optional[int]): Width of the generated images.
+            height (Optional[int]): Height of the generated images.
+            width (Optional[int]): Width of the generated images.
+            sample_size (Optional[Tuple[int, int]]): Spatial dimensions for the UNet model.
+            image_size (Optional[Tuple[int, int]]): Alternative way to specify image dimensions.
+                Cannot be used together with img_height/img_width.
+            guidance_scale (Optional[float]): Scale for classifier-free guidance.
+            kwargs: Additional arguments passed to the parent RBLNModelConfig.
+
+        Raises:
+            ValueError: If both image_size and img_height/img_width are provided.
+
+        Note:
+            When guidance_scale > 1.0, the UNet batch size is automatically doubled to
+            accommodate classifier-free guidance.
+        """
+        super().__init__(**kwargs)
+
+        # Initial check for image_size conflict remains as is
+        if image_size is not None and (
+            img_height is not None or img_width is not None or height is not None or width is not None
+        ):
+            raise ValueError("image_size cannot be provided alongside img_height/img_width or height/width")
+
+        # Prioritize height/width (HF-aligned)
+        if height is not None and width is not None:
+            if img_height is not None or img_width is not None:
+                # Raise error if both sets of arguments are provided
+                raise ValueError(
+                    "Cannot provide both 'height'/'width' and 'img_height'/'img_width' simultaneously. "
+                    "Please use one set of arguments for image dimensions, preferring 'height'/'width'."
+                )
+            image_size = (height, width)
+        elif (height is not None and width is None) or (height is None and width is not None):
+            raise ValueError("Both height and width must be provided together if used")
+        # Fallback to img_height/img_width for backward compatibility
+        elif img_height is not None and img_width is not None:
+            image_size = (img_height, img_width)
+        elif (img_height is not None and img_width is None) or (img_height is None and img_width is not None):
+            raise ValueError("Both img_height and img_width must be provided together if used")
+
+        self.text_encoder = self.initialize_submodule_config(
+            text_encoder,
+            cls_name="RBLNCLIPTextModelConfig",
+            batch_size=batch_size,
+        )
+        self.text_encoder_2 = self.initialize_submodule_config(
+            text_encoder_2,
+            cls_name="RBLNCLIPTextModelWithProjectionConfig",
+            batch_size=batch_size,
+        )
+
+        self.unet = self.initialize_submodule_config(
+            unet,
+            cls_name="RBLNUNet2DConditionModelConfig",
+            sample_size=sample_size,
+        )
+        self.vae = self.initialize_submodule_config(
+            vae,
+            cls_name="RBLNAutoencoderKLConfig",
+            batch_size=batch_size,
+            uses_encoder=self.__class__._vae_uses_encoder,
+            sample_size=image_size,  # image size is equal to sample size in vae
+        )
+
+        # Get default guidance scale from original class to set UNet batch size
+        if guidance_scale is None:
+            guidance_scale = self.get_default_values_for_original_cls("__call__", ["guidance_scale"])["guidance_scale"]
+
+        if not self.unet.batch_size_is_specified:
+            do_classifier_free_guidance = guidance_scale > 1.0
+            if do_classifier_free_guidance:
+                self.unet.batch_size = self.text_encoder.batch_size * 2
+            else:
+                self.unet.batch_size = self.text_encoder.batch_size
+
+    @property
+    def batch_size(self):
+        return self.vae.batch_size
+
+    @property
+    def sample_size(self):
+        return self.unet.sample_size
+
+    @property
+    def image_size(self):
+        return self.vae.sample_size
+
+
+class RBLNStableDiffusionXLPipelineConfig(RBLNStableDiffusionXLPipelineBaseConfig):
+    """Config for SDXL Text2Img Pipeline"""
+
+    _vae_uses_encoder = False
+
+
+class RBLNStableDiffusionXLImg2ImgPipelineConfig(RBLNStableDiffusionXLPipelineBaseConfig):
+    """Config for SDXL Img2Img Pipeline"""
+
+    _vae_uses_encoder = True
+
+
+class RBLNStableDiffusionXLInpaintPipelineConfig(RBLNStableDiffusionXLPipelineBaseConfig):
+    """Config for SDXL Inpainting Pipeline"""
+
+    _vae_uses_encoder = True

optimum/rbln/diffusers/configurations/pipelines/configuration_stable_video_diffusion.py

@@ -0,0 +1,114 @@
+# Copyright 2025 Rebellions Inc. 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 typing import Any, Optional
+
+from ....configuration_utils import RBLNModelConfig
+from ....transformers import RBLNCLIPVisionModelWithProjectionConfig
+from ..models import RBLNAutoencoderKLTemporalDecoderConfig, RBLNUNetSpatioTemporalConditionModelConfig
+
+
+class RBLNStableVideoDiffusionPipelineConfig(RBLNModelConfig):
+    submodules = ["image_encoder", "unet", "vae"]
+    _vae_uses_encoder = True
+
+    def __init__(
+        self,
+        image_encoder: Optional[RBLNCLIPVisionModelWithProjectionConfig] = None,
+        unet: Optional[RBLNUNetSpatioTemporalConditionModelConfig] = None,
+        vae: Optional[RBLNAutoencoderKLTemporalDecoderConfig] = None,
+        *,
+        batch_size: Optional[int] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_frames: Optional[int] = None,
+        decode_chunk_size: Optional[int] = None,
+        guidance_scale: Optional[float] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            image_encoder (Optional[RBLNCLIPVisionModelWithProjectionConfig]): Configuration for the image encoder component.
+                Initialized as RBLNCLIPVisionModelWithProjectionConfig if not provided.
+            unet (Optional[RBLNUNetSpatioTemporalConditionModelConfig]): Configuration for the UNet model component.
+                Initialized as RBLNUNetSpatioTemporalConditionModelConfig if not provided.
+            vae (Optional[RBLNAutoencoderKLTemporalDecoderConfig]): Configuration for the VAE model component.
+                Initialized as RBLNAutoencoderKLTemporalDecoderConfig if not provided.
+            batch_size (Optional[int]): Batch size for inference, applied to all submodules.
+            height (Optional[int]): Height of the generated images.
+            width (Optional[int]): Width of the generated images.
+            num_frames (Optional[int]): The number of frames in the generated video.
+            decode_chunk_size (Optional[int]): The number of frames to decode at once during VAE decoding.
+                Useful for managing memory usage during video generation.
+            guidance_scale (Optional[float]): Scale for classifier-free guidance.
+            kwargs: Additional arguments passed to the parent RBLNModelConfig.
+
+        Raises:
+            ValueError: If both image_size and height/width are provided.
+
+        Note:
+            When guidance_scale > 1.0, the UNet batch size is automatically doubled to
+            accommodate classifier-free guidance.
+        """
+        super().__init__(**kwargs)
+        if height is not None and width is not None:
+            image_size = (height, width)
+        else:
+            # Get default image size from original class to set UNet, VAE image size
+            height = self.get_default_values_for_original_cls("__call__", ["height"])["height"]
+            width = self.get_default_values_for_original_cls("__call__", ["width"])["width"]
+            image_size = (height, width)
+
+        self.image_encoder = self.initialize_submodule_config(
+            image_encoder, cls_name="RBLNCLIPVisionModelWithProjectionConfig", batch_size=batch_size
+        )
+        self.unet = self.initialize_submodule_config(
+            unet,
+            cls_name="RBLNUNetSpatioTemporalConditionModelConfig",
+            num_frames=num_frames,
+        )
+        self.vae = self.initialize_submodule_config(
+            vae,
+            cls_name="RBLNAutoencoderKLTemporalDecoderConfig",
+            batch_size=batch_size,
+            num_frames=num_frames,
+            decode_chunk_size=decode_chunk_size,
+            uses_encoder=self.__class__._vae_uses_encoder,
+            sample_size=image_size,  # image size is equal to sample size in vae
+        )
+
+        # Get default guidance scale from original class to set UNet batch size
+        if guidance_scale is None:
+            guidance_scale = self.get_default_values_for_original_cls("__call__", ["max_guidance_scale"])[
+                "max_guidance_scale"
+            ]
+
+        if not self.unet.batch_size_is_specified:
+            do_classifier_free_guidance = guidance_scale > 1.0
+            if do_classifier_free_guidance:
+                self.unet.batch_size = self.image_encoder.batch_size * 2
+            else:
+                self.unet.batch_size = self.image_encoder.batch_size
+
+    @property
+    def batch_size(self):
+        return self.vae.batch_size
+
+    @property
+    def sample_size(self):
+        return self.unet.sample_size
+
+    @property
+    def image_size(self):
+        return self.vae.sample_size