Upload 2 files

handler.py

@@ -0,0 +1,62 @@
+from typing import Dict, List, Any
+from PIL import Image
+from transformers import Pix2StructProcessor, Pix2StructForConditionalGeneration
+import torch
+import base64
+import io
+
+class EndpointHandler:
+    def __init__(self, path: str = ""):
+        """Called when the endpoint starts. Load model and processor."""
+        self.processor = Pix2StructProcessor.from_pretrained(path)
+        self.model = Pix2StructForConditionalGeneration.from_pretrained(path)
+        
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.model.to(self.device)
+        self.model.eval()
+
+    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
+        """
+        Called on every request.
+        
+        Args:
+            data: Dictionary containing:
+                - inputs: base64 encoded image string
+                - parameters (optional): generation params like max_new_tokens
+        
+        Returns:
+            List containing the generated table text
+        """
+        inputs = data.get("inputs")
+        parameters = data.get("parameters", {})
+        
+        # Decode base64 image
+        if isinstance(inputs, str):
+            image_bytes = base64.b64decode(inputs)
+            image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
+        else:
+            raise ValueError("Expected base64 encoded image string in 'inputs'")
+        
+        # Process image
+        model_inputs = self.processor(
+            images=image,
+            return_tensors="pt"
+        ).to(self.device)
+        
+        # Get generation parameters
+        max_new_tokens = parameters.get("max_new_tokens", 512)
+        
+        # Generate
+        with torch.no_grad():
+            predictions = self.model.generate(
+                **model_inputs,
+                max_new_tokens=max_new_tokens
+            )
+        
+        # Decode
+        output_text = self.processor.decode(
+            predictions[0],
+            skip_special_tokens=True
+        )
+        
+        return [{"generated_text": output_text}]

requirements.txt

@@ -0,0 +1,4 @@
+transformers>=4.30.0
+torch
+Pillow
+sentencepiece