Update app.py

app.py

@@ -5,15 +5,12 @@
 import os
 import re
 import gradio as gr
-import dashscope
-from dashscope import MultiModalConversation
+import base64
+from huggingface_hub import InferenceClient
 from argparse import ArgumentParser
-from http import HTTPStatus
-from urllib3.exceptions import HTTPError
 
 # Set API key
-API_KEY = os.environ['API_KEY']
-dashscope.api_key = API_KEY
+client = InferenceClient(provider="hf-inference")
 
 # Constants
 REVISION = 'v1.0.4'
@@ -30,6 +27,11 @@ def _get_args():
                         help="Demo server name.")
     return parser.parse_args()
 
+def image_to_base64(image_path):
+    """Convert an image file to base64 string."""
+    with open(image_path, "rb") as image_file:
+        return base64.b64encode(image_file.read()).decode('utf-8')
+
 def parse_bounding_boxes(text):
     """Parse bounding box coordinates from model output."""
     matches = re.findall(BOX_TAG_PATTERN, text)
@@ -44,28 +46,34 @@ def predict(image, prompt):
     if image is None or not prompt:
         return None, "Please upload an image and provide a prompt."
 
-    # Prepare message for the model
-    messages = [{
-        'role': 'user',
-        'content': [
-            {'image': f'file://{image}'},
-            {'text': prompt}
+    try:
+        # Convert image to base64
+        image_base64 = image_to_base64(image)
+        image_url = f"data:image/jpeg;base64,{image_base64}"
+
+        # Prepare message for the model
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": prompt},
+                    {"type": "image_url", "image_url": {"url": image_url}}
+                ]
+            }
         ]
-    }]
 
-    # Call the Qwen2.5-VL model
-    try:
-        responses = MultiModalConversation.call(
-            model='qwen2.5-vl-32b-instruct',
+        # Call the Hugging Face Inference API
+        stream = client.chat.completions.create(
+            model="Qwen/Qwen2.5-VL-32B-Instruct",
             messages=messages,
-            stream=False
+            stream=True
         )
-        if responses.status_code != HTTPStatus.OK:
-            return None, f"Error: {responses.message}"
 
-        # Extract response text
-        response = responses.output.choices[0].message.content
-        response_text = ''.join([ele['text'] if 'text' in ele else ele.get('box', '') for ele in response])
+        # Aggregate streaming response
+        response_text = ""
+        for chunk in stream:
+            if chunk.choices[0].delta.content:
+                response_text += chunk.choices[0].delta.content
 
         # Parse bounding boxes
         bboxes = parse_bounding_boxes(response_text)
@@ -75,8 +83,6 @@ def predict(image, prompt):
         # Return the image and annotations for AnnotatedImage
         return (image, bboxes), None
 
-    except HTTPError as e:
-        return None, f"HTTP Error: {str(e)}"
     except Exception as e:
         return None, f"Error: {str(e)}"
 
@@ -86,7 +92,7 @@ def clear_inputs():
 
 def _launch_demo(args):
     with gr.Blocks() as demo:
-        gr.Markdown("""<center><font size=3> Qwen2.5-VL-32B-Instruct Bounding Box Demo </center>""")
+        gr.Markdown("""<center><font size=3> Qwen2.5-VL-32B-Instruct Bounding Box Demo (Hugging Face) </center>""")
 
         with gr.Row():
             with gr.Column():