optimum-rbln 0.7.5rc1__tar.gz → 0.7.5rc3__tar.gz

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/.github/workflows/rbln_optimum_pytest.yaml

@@ -25,6 +25,7 @@ env:
   REBEL_PYPI_ENDPOINT: ${{ vars.REBEL_PYPI_INTERNAL_ENDPOINT }}
   REBEL_PYPI_USERNAME: ${{ secrets.REBEL_PYPI_USERNAME }}
   REBEL_PYPI_PASSWORD: ${{ secrets.REBEL_PYPI_PASSWORD }}
+  HF_HOME: ${{ secrets.HF_HOME }}
 
 jobs:
   pytest:

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: optimum-rbln
-Version: 0.7.5rc1
+Version: 0.7.5rc3
 Summary: Optimum RBLN is the interface between the HuggingFace Transformers and Diffusers libraries and RBLN accelerators. It provides a set of tools enabling easy model loading and inference on single and multiple rbln device settings for different downstream tasks.
 Project-URL: Homepage, https://rebellions.ai
 Project-URL: Documentation, https://docs.rbln.ai

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/__version__.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.7.5rc1'
-__version_tuple__ = version_tuple = (0, 7, 5, 'rc1')
+__version__ = version = '0.7.5rc3'
+__version_tuple__ = version_tuple = (0, 7, 5, 'rc3')

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/diffusers/configurations/pipelines/configuration_controlnet.py

@@ -33,6 +33,8 @@ class _RBLNStableDiffusionControlNetPipelineBaseConfig(RBLNModelConfig):
         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,
@@ -51,6 +53,8 @@ class _RBLNStableDiffusionControlNetPipelineBaseConfig(RBLNModelConfig):
             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.
@@ -65,11 +69,29 @@ class _RBLNStableDiffusionControlNetPipelineBaseConfig(RBLNModelConfig):
             accommodate classifier-free guidance.
         """
         super().__init__(**kwargs)
-        if image_size is not None and (img_height is not None or img_width is not None):
-            raise ValueError("image_size and img_height/img_width cannot both be provided")
 
-        if img_height is not None and img_width is not None:
+        # 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.init_submodule_config(RBLNCLIPTextModelConfig, text_encoder, batch_size=batch_size)
         self.unet = self.init_submodule_config(
@@ -139,6 +161,8 @@ class _RBLNStableDiffusionXLControlNetPipelineBaseConfig(RBLNModelConfig):
         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,
@@ -159,6 +183,8 @@ class _RBLNStableDiffusionXLControlNetPipelineBaseConfig(RBLNModelConfig):
             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.
@@ -173,11 +199,29 @@ class _RBLNStableDiffusionXLControlNetPipelineBaseConfig(RBLNModelConfig):
             accommodate classifier-free guidance.
         """
         super().__init__(**kwargs)
-        if image_size is not None and (img_height is not None or img_width is not None):
-            raise ValueError("image_size and img_height/img_width cannot both be provided")
 
-        if img_height is not None and img_width is not None:
+        # 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.init_submodule_config(RBLNCLIPTextModelConfig, text_encoder, batch_size=batch_size)
         self.text_encoder_2 = self.init_submodule_config(

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/diffusers/configurations/pipelines/configuration_kandinsky2_2.py

@@ -35,6 +35,8 @@ class _RBLNKandinskyV22PipelineBaseConfig(RBLNModelConfig):
         image_size: Optional[Tuple[int, int]] = None,
         img_height: Optional[int] = None,
         img_width: Optional[int] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         **kwargs,
     ):
         """
@@ -50,6 +52,8 @@ class _RBLNKandinskyV22PipelineBaseConfig(RBLNModelConfig):
                 Cannot be used together with img_height/img_width.
             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.
             **kwargs: Additional arguments passed to the parent RBLNModelConfig.
 
         Raises:
@@ -60,11 +64,29 @@ class _RBLNKandinskyV22PipelineBaseConfig(RBLNModelConfig):
             accommodate classifier-free guidance.
         """
         super().__init__(**kwargs)
-        if image_size is not None and (img_height is not None or img_width is not None):
-            raise ValueError("image_size and img_height/img_width cannot both be provided")
 
-        if img_height is not None and img_width is not None:
+        # 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.unet = self.init_submodule_config(RBLNUNet2DConditionModelConfig, unet, sample_size=sample_size)
         self.movq = self.init_submodule_config(
@@ -186,6 +208,8 @@ class _RBLNKandinskyV22CombinedPipelineBaseConfig(RBLNModelConfig):
         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,
         prior_prior: Optional[RBLNPriorTransformerConfig] = None,
         prior_image_encoder: Optional[RBLNCLIPVisionModelWithProjectionConfig] = None,
@@ -212,6 +236,8 @@ class _RBLNKandinskyV22CombinedPipelineBaseConfig(RBLNModelConfig):
             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.
             prior_prior (Optional[RBLNPriorTransformerConfig]): Direct configuration for the prior transformer.
                 Used if prior_pipe is not provided.
@@ -226,6 +252,30 @@ class _RBLNKandinskyV22CombinedPipelineBaseConfig(RBLNModelConfig):
             **kwargs: Additional arguments passed to the parent RBLNModelConfig.
         """
         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.prior_pipe = self.init_submodule_config(
             RBLNKandinskyV22PriorPipelineConfig,
             prior_pipe,
@@ -243,8 +293,6 @@ class _RBLNKandinskyV22CombinedPipelineBaseConfig(RBLNModelConfig):
             batch_size=batch_size,
             sample_size=sample_size,
             image_size=image_size,
-            img_height=img_height,
-            img_width=img_width,
             guidance_scale=guidance_scale,
         )
 

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion.py

@@ -32,6 +32,8 @@ class _RBLNStableDiffusionPipelineBaseConfig(RBLNModelConfig):
         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,
@@ -48,6 +50,8 @@ class _RBLNStableDiffusionPipelineBaseConfig(RBLNModelConfig):
             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.
@@ -62,11 +66,29 @@ class _RBLNStableDiffusionPipelineBaseConfig(RBLNModelConfig):
             accommodate classifier-free guidance.
         """
         super().__init__(**kwargs)
-        if image_size is not None and (img_height is not None or img_width is not None):
-            raise ValueError("image_size and img_height/img_width cannot both be provided")
 
-        if img_height is not None and img_width is not None:
+        # 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.init_submodule_config(RBLNCLIPTextModelConfig, text_encoder, batch_size=batch_size)
         self.unet = self.init_submodule_config(

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_3.py

@@ -37,6 +37,8 @@ class _RBLNStableDiffusion3PipelineBaseConfig(RBLNModelConfig):
         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,
     ):
@@ -59,6 +61,8 @@ class _RBLNStableDiffusion3PipelineBaseConfig(RBLNModelConfig):
             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.
 
@@ -70,11 +74,29 @@ class _RBLNStableDiffusion3PipelineBaseConfig(RBLNModelConfig):
             accommodate classifier-free guidance.
         """
         super().__init__(**kwargs)
-        if image_size is not None and (img_height is not None or img_width is not None):
-            raise ValueError("image_size and img_height/img_width cannot both be provided")
 
-        if img_height is not None and img_width is not None:
+        # 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
 

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/diffusers/configurations/pipelines/configuration_stable_diffusion_xl.py

@@ -33,6 +33,8 @@ class _RBLNStableDiffusionXLPipelineBaseConfig(RBLNModelConfig):
         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,
@@ -51,6 +53,8 @@ class _RBLNStableDiffusionXLPipelineBaseConfig(RBLNModelConfig):
             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.
@@ -65,11 +69,29 @@ class _RBLNStableDiffusionXLPipelineBaseConfig(RBLNModelConfig):
             accommodate classifier-free guidance.
         """
         super().__init__(**kwargs)
-        if image_size is not None and (img_height is not None or img_width is not None):
-            raise ValueError("image_size and img_height/img_width cannot both be provided")
 
-        if img_height is not None and img_width is not None:
+        # 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.init_submodule_config(RBLNCLIPTextModelConfig, text_encoder, batch_size=batch_size)
         self.text_encoder_2 = self.init_submodule_config(

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/modeling.py

@@ -282,7 +282,7 @@ class RBLNModel(RBLNBaseModel):
         Prepare model output based on return_dict flag.
         This method can be overridden by subclasses to provide task-specific output handling.
         """
-        tuple_output = (output,) if not isinstance(output, (tuple, list)) else output
+        tuple_output = (output,) if not isinstance(output, (tuple, list)) else tuple(output)
         if not return_dict:
             return tuple_output
         else:

{optimum_rbln-0.7.5rc1 → optimum_rbln-0.7.5rc3}/src/optimum/rbln/transformers/models/blip_2/configuration_blip_2.py

@@ -18,29 +18,12 @@ from ....configuration_utils import RBLNModelConfig
 
 
 class RBLNBlip2VisionModelConfig(RBLNModelConfig):
-    def __init__(
-        self,
-        batch_size: Optional[int] = None,
-        **kwargs,
-    ):
-        """
-        Args:
-            batch_size (Optional[int]): The batch size for inference. Defaults to 1.
-            **kwargs: Additional arguments passed to the parent RBLNModelConfig.
-
-        Raises:
-            ValueError: If batch_size is not a positive integer.
-        """
-        super().__init__(**kwargs)
-        self.batch_size = batch_size or 1
-        if not isinstance(self.batch_size, int) or self.batch_size < 0:
-            raise ValueError(f"batch_size must be a positive integer, got {self.batch_size}")
+    pass
 
 
 class RBLNBlip2QFormerModelConfig(RBLNModelConfig):
     def __init__(
         self,
-        batch_size: Optional[int] = None,
         num_query_tokens: Optional[int] = None,
         image_text_hidden_size: Optional[int] = None,
         **kwargs,
@@ -54,10 +37,6 @@ class RBLNBlip2QFormerModelConfig(RBLNModelConfig):
             ValueError: If batch_size is not a positive integer.
         """
         super().__init__(**kwargs)
-        self.batch_size = batch_size or 1
-        if not isinstance(self.batch_size, int) or self.batch_size < 0:
-            raise ValueError(f"batch_size must be a positive integer, got {self.batch_size}")
-
         self.num_query_tokens = num_query_tokens
         self.image_text_hidden_size = image_text_hidden_size
 
@@ -88,6 +67,6 @@ class RBLNBlip2ForConditionalGenerationConfig(RBLNModelConfig):
         if not isinstance(self.batch_size, int) or self.batch_size < 0:
             raise ValueError(f"batch_size must be a positive integer, got {self.batch_size}")
 
-        self.vision_model = self.init_submodule_config(RBLNBlip2VisionModelConfig, vision_model, batch_size=batch_size)
+        self.vision_model = self.init_submodule_config(RBLNBlip2VisionModelConfig, vision_model)
         self.language_model = language_model
-        self.qformer = self.init_submodule_config(RBLNBlip2QFormerModelConfig, qformer, batch_size=batch_size)
+        self.qformer = self.init_submodule_config(RBLNBlip2QFormerModelConfig, qformer)