Update handler.py

handler.py

@@ -1,31 +1,110 @@
-from typing import Dict, List, Any
-from lmdeploy import pipeline
-from lmdeploy.vl import load_image
-from lmdeploy.messages import TurbomindEngineConfig
+from transformers import AutoTokenizer, AutoModel
+import torch
+import torchvision.transforms as T
+from PIL import Image
+from io import BytesIO
+import requests
+from torchvision.transforms.functional import InterpolationMode
 
 class EndpointHandler():
     def __init__(self, path):
-        # Preload the model at initialization
-        backend_config = TurbomindEngineConfig(model_name ="OpenGVLab/InternVL-Chat-V1-5",model_format='hf',tp=1)
-        self.pipe = pipeline(f"{path}", backend_config=backend_config, log_level='INFO')
-
-    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
-        """
-        data args:
-            inputs (:obj: `str` | `PIL.Image` | `np.array`)
-            kwargs
-        Return:
-            A :obj:`list` | `dict`: will be serialized and returned
-        """
+
+        # If you have an 80G A100 GPU, you can put the entire model on a single GPU.
+        self.model = AutoModel.from_pretrained(
+            path,
+            torch_dtype=torch.bfloat16,
+            low_cpu_mem_usage=True,
+            trust_remote_code=True).eval().cuda()
+
+        self.tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True)
+        self.IMAGENET_MEAN = (0.485, 0.456, 0.406)
+        self.IMAGENET_STD = (0.229, 0.224, 0.225)
+
+    def __call__(self, data):
+
         inputs = data.pop("inputs", data)
         image_url = inputs.get("image_url")
         prompt = inputs.get("prompt")
-        
-        if not image_url:
-            return [{'error': 'No image URL provided'}]
-
-        image = load_image(image_url)
-       
-        
-        response = self.pipe((prompt, image))
-        return {'response': response.text}
+        # set the max number of tiles in `max_num`
+        pixel_values = self.load_image(image_url, max_num=6).to(torch.bfloat16).cuda()
+        generation_config = dict(
+            num_beams=1,
+            max_new_tokens=1000,
+            do_sample=False,
+        )
+
+        # single-round single-image conversation
+        response = self.model.chat(self.tokenizer, pixel_values, prompt, generation_config)
+        return {"response": response}
+
+    def load_image(self, image_file, input_size=448, max_num=6):
+        response = requests.get(image_file)
+        image = Image.open(BytesIO(response.content)).convert('RGB')
+        transform = self.build_transform(input_size=input_size)
+        images = self.dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
+        pixel_values = [transform(image) for image in images]
+        pixel_values = torch.stack(pixel_values)
+        return pixel_values
+
+    def build_transform(self, input_size):
+        MEAN, STD = self.IMAGENET_MEAN, self.IMAGENET_STD
+        transform = T.Compose([
+            T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
+            T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
+            T.ToTensor(),
+            T.Normalize(mean=MEAN, std=STD)
+        ])
+        return transform
+
+    def dynamic_preprocess(self, image, min_num=1, max_num=6, image_size=448, use_thumbnail=False):
+        orig_width, orig_height = image.size
+        aspect_ratio = orig_width / orig_height
+
+        # calculate the existing image aspect ratio
+        target_ratios = set(
+            (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
+            i * j <= max_num and i * j >= min_num)
+        target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
+
+        # find the closest aspect ratio to the target
+        target_aspect_ratio = self.find_closest_aspect_ratio(
+            aspect_ratio, target_ratios, orig_width, orig_height, image_size)
+
+        # calculate the target width and height
+        target_width = image_size * target_aspect_ratio[0]
+        target_height = image_size * target_aspect_ratio[1]
+        blocks = target_aspect_ratio[0] * target_aspect_ratio[1]
+
+        # resize the image
+        resized_img = image.resize((target_width, target_height))
+        processed_images = []
+        for i in range(blocks):
+            box = (
+                (i % (target_width // image_size)) * image_size,
+                (i // (target_width // image_size)) * image_size,
+                ((i % (target_width // image_size)) + 1) * image_size,
+                ((i // (target_width // image_size)) + 1) * image_size
+            )
+            # split the image
+            split_img = resized_img.crop(box)
+            processed_images.append(split_img)
+        assert len(processed_images) == blocks
+        if use_thumbnail and len(processed_images) != 1:
+            thumbnail_img = image.resize((image_size, image_size))
+            processed_images.append(thumbnail_img)
+        return processed_images
+
+    def find_closest_aspect_ratio(self, aspect_ratio, target_ratios, width, height, image_size):
+        best_ratio_diff = float('inf')
+        best_ratio = (1, 1)
+        area = width * height
+        for ratio in target_ratios:
+            target_aspect_ratio = ratio[0] / ratio[1]
+            ratio_diff = abs(aspect_ratio - target_aspect_ratio)
+            if ratio_diff < best_ratio_diff:
+                best_ratio_diff = ratio_diff
+                best_ratio = ratio
+            elif ratio_diff == best_ratio_diff:
+                if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
+                    best_ratio = ratio
+        return best_ratio