sd-inpainting: pass prompt in kwargs & reduce GPU usage

Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/11616099

* pass prompt in kwargs

modelscope/pipelines/cv/image_inpainting_sdv2_pipeline.py

@@ -56,13 +56,29 @@ class ImageInpaintingSDV2Pipeline(DiffusersPipeline):
         """
         super().__init__(model, device, **kwargs)
 
-        torch_dtype = kwargs.get('torch_dtype', torch.float32)
+        torch_dtype = kwargs.get('torch_dtype', torch.float16)
 
         # build upon the diffuser stable diffusion pipeline
         self.pipeline = StableDiffusionInpaintPipeline.from_pretrained(
             model, torch_dtype=torch_dtype)
         self.pipeline.to(self.device)
 
+        enable_attention_slicing = kwargs.get('enable_attention_slicing', True)
+        if enable_attention_slicing:
+            self.pipeline.enable_attention_slicing()
+
+    def _sanitize_parameters(self, **pipeline_parameters):
+        """
+        this method should sanitize the keyword args to preprocessor params,
+        forward params and postprocess params on '__call__' or '_process_single' method
+
+        Returns:
+            Dict[str, str]:  preprocess_params = {}
+            Dict[str, str]:  forward_params = pipeline_parameters
+            Dict[str, str]:  postprocess_params = pipeline_parameters
+        """
+        return {}, pipeline_parameters, pipeline_parameters
+
     def forward(self, inputs: Dict[str, Any],
                 **forward_params) -> Dict[str, Any]:
         if not isinstance(inputs, dict):
@@ -76,13 +92,20 @@ class ImageInpaintingSDV2Pipeline(DiffusersPipeline):
         num_images_per_prompt = inputs.get('num_images_per_prompt', 1)
         eta = inputs.get('eta', 0.0)
 
+        if 'prompt' in inputs.keys():
+            prompt = inputs['prompt']
+        else:
+            # for demo_service
+            prompt = forward_params.get('prompt', 'background')
+        print(f'Test with prompt: {prompt}')
+
         image = load_image(inputs['image'])
         mask = load_image(inputs['mask'])
-        prompt = inputs['prompt']
+
         w, h = image.size
         print(f'loaded input image of size ({w}, {h})')
         width, height = map(lambda x: x - x % 64,
-                            (w, h))  # resize to integer multiple of 32
+                            (w, h))  # resize to integer multiple of 64
         image = image.resize((width, height))
         mask = mask.resize((width, height))
         out_image = self.pipeline(

requirements/cv.txt

@@ -54,7 +54,7 @@ timm>=0.4.9
 torchmetrics>=0.6.2
 torchsummary>=1.5.1
 torchvision
-transformers>=4.19.2
+transformers>=4.26.0
 ujson
 utils
 videofeatures_clipit>=1.0