Upload 3 files

handler.py

@@ -0,0 +1,56 @@
+import base64
+from io import BytesIO
+from typing import  Dict, List, Any
+from transformers import Pix2StructForConditionalGeneration, AutoProcessor
+from PIL import Image
+import torch
+
+class EndpointHandler():
+
+    def __init__(self):
+
+        model_name = "google/pix2struct-infographics-vqa-large"
+
+
+        self.model = Pix2StructForConditionalGeneration.from_pretrained(model_name)
+        self.processor = AutoProcessor.from_pretrained(model_name)
+        self.text_prompt = None # 
+
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+        self.model.to(self.device)
+
+
+    def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
+        """
+        Args:
+            data (:obj:):
+                includes the input data and the parameters for the inference.
+        Return:
+            a dictionary with the output of the model. The only key is `output` and the 
+            value is a list of str.            
+        """
+        inputs = data.pop("inputs", data)
+        parameters = data.pop("parameters", {})
+
+        if isinstance(inputs["image"], list):
+            img = [Image.open(BytesIO(base64.b64decode(img))) for img in inputs['image']]
+        else:
+            img = Image.open(BytesIO(base64.b64decode(inputs['image'])))
+
+        question = inputs['question']
+
+
+        
+        with torch.inference_mode():
+            model_inputs = self.processor(images=img, text=question, return_tensors="pt").to(self.device)
+
+            raw_output = self.model.generate(**model_inputs, **parameters)
+
+            decoded_output = self.processor.batch_decode(raw_output, skip_special_tokens=True)
+
+
+        # postprocess the prediction
+        return {
+            "output": decoded_output
+        }

requirements.txt

@@ -0,0 +1,2 @@
+transformers==4.35.1
+sentencepiece==0.1.99

test.ipynb

@@ -0,0 +1,88 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import handler\n",
+    "import importlib\n",
+    "\n",
+    "importlib.reload(handler)\n",
+    "\n",
+    "import handler\n",
+    "\n",
+    "h = handler.EndpointHandler()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import base64\n",
+    "from pathlib import Path\n",
+    "\n",
+    "if not Path(\"What-is-an-infographic.jpg\").exists():\n",
+    "    !wget https://visme.co/blog/wp-content/uploads/2020/02/What-is-an-infographic.jpg\n",
+    "\n",
+    "with open(\"What-is-an-infographic.jpg\", \"rb\") as f:\n",
+    "    b64 = base64.b64encode(f.read())\n",
+    "\n",
+    "question = \"What percent of information do we understand through body language?\"\n",
+    "\n",
+    "payload = {\n",
+    "    \"inputs\": {\n",
+    "        \"image\": [b64.decode(\"utf-8\")]*2, \n",
+    "        \"question\": [question]*2\n",
+    "        }, \n",
+    "    \"parameters\":{\n",
+    "        \"max_new_tokens\": 10,\n",
+    "    }}"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'output': ['55%', '55%']}"
+      ]
+     },
+     "execution_count": 14,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "h(payload)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.12"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}