Update app.py

app.py

@@ -25,23 +25,29 @@ def model_inference(
     user_content = []
     media_queue = []
     if history == []:
-        for file in input_dict["files"]:
+        text = input_dict["text"].strip() 
+        
+        for file in input_dict.get("files", []):
             if file.endswith((".png", ".jpg", ".jpeg", ".gif", ".bmp")):
                 media_queue.append({"type": "image", "path": file})
             elif file.endswith((".mp4", ".mov", ".avi", ".mkv", ".flv")):
                 media_queue.append({"type": "video", "path": file})
 
-        text = input_dict.get("text", "")
-        parts = re.split(r'(<image>|<video>)', text)  
+        if "<image>" in text or "<video>" in text:
+            parts = re.split(r'(<image>|<video>)', text)  
+            for part in parts:
+                if part == "<image>" and media_queue:
+                    user_content.append(media_queue.pop(0)) 
+                elif part == "<video>" and media_queue:
+                    user_content.append(media_queue.pop(0))  
+                elif part.strip():  
+                    user_content.append({"type": "text", "text": part.strip()})
+        else:
+            user_content.append({"type": "text", "text": text})
+            
+            for media in media_queue:
+                user_content.append(media)
 
-        for part in parts:
-            if part == "<image>" and media_queue:
-                user_content.append(media_queue.pop(0)) 
-            elif part == "<video>" and media_queue:
-                user_content.append(media_queue.pop(0)) 
-            elif part.strip():  
-                user_content.append({"type": "text", "text": part.strip()})
-        
         resulting_messages = [{"role": "user", "content": user_content}]
 
     elif len(history) > 0:
@@ -51,7 +57,7 @@ def model_inference(
         for hist in history:
             if hist["role"] == "user" and isinstance(hist["content"], tuple): 
                 file_name = hist["content"][0]
-            if file_name.endswith((".png", ".jpg", ".jpeg", ".gif", ".bmp")):
+            if file_name.endswith((".png", ".jpg", ".jpeg")):
                 media_queue.append({"type": "image", "path": file_name})
             elif file_name.endswith(".mp4"):
                 media_queue.append({"type": "video", "path": file_name})
@@ -120,19 +126,15 @@ def model_inference(
 
 
 examples=[
-              [{"text": "Can you describe this image?", "files": ["example_images/newyork.jpg"]}],
-              [{"text": "Can you describe this image?", "files": ["example_images/dogs.jpg"]}],
-              [{"text": "Where do the severe droughts happen according to this diagram?", "files": ["example_images/examples_weather_events.png"]}],
+              [[{"text": "Where do the severe droughts happen according to this diagram?", "files": ["example_images/examples_weather_events.png"]}],
               [{"text": "What art era this artpiece <image> and this artpiece <image> belong to?", "files": ["example_images/rococo.jpg", "example_images/rococo_1.jpg"]}],
               [{"text": "Describe this image.", "files": ["example_images/campeones.jpg"]}],
               [{"text": "What does this say?", "files": ["example_images/math.jpg"]}],
               [{"text": "What is the date in this document?", "files": ["example_images/document.jpg"]}],
-              [{"text": "What is this UI about?", "files": ["example_images/s2w_example.png"]}],
-            [{"text": "What is happening in the video?", "files": ["example_images/barcamadridhighlights.mp4"]}],
-
+              [{"text": "What is happening in the video?", "files": ["example_images/barcamadridhighlights.mp4"]}],
       ]
 demo = gr.ChatInterface(fn=model_inference, title="SmolVLM2: The Smollest Video Model Ever 📺", 
-                description="Play with [SmolVLM2-2.2B-Instruct](https://huggingface.co/HuggingFaceTB/SmolVLM2-2.2B-Instruct) in this demo. To get started, upload an image and text or try one of the examples. To see how to interleave images, check the multiple image example.",
+                description="Play with [SmolVLM2-2.2B-Instruct](https://huggingface.co/HuggingFaceTB/SmolVLM2-2.2B-Instruct) in this demo. To get started, upload an image and text or try one of the examples. This demo doesn't use history for the chat, so every chat you start is a new conversation.",
                 examples=examples,
                 textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image", ".mp4"], file_count="multiple"), stop_btn="Stop Generation", multimodal=True,
                 cache_examples=False,