Add some limits to uploaded files and a max to send to the model

app.py

@@ -40,6 +40,9 @@ openai_client = None
 USE_RANDOM_ENDPOINT = False
 endpoint_rotation_count = 0
 
+# Maximum number of image messages allowed per request
+MAX_IMAGE_MESSAGES = 5
+
 
 def app_loaded(state, request: gr.Request):
     message_html = setup_model(DEFAULT_MODEL_NAME, intial=False)
@@ -160,6 +163,11 @@ def run_chat_inference(history, message, state):
             yield history, INPUT_ENABLED, SEND_BUTTON_ENABLED, STOP_BUTTON_DISABLED, BUTTON_ENABLED, state
             return history, INPUT_ENABLED, SEND_BUTTON_ENABLED, STOP_BUTTON_DISABLED, BUTTON_ENABLED, state
 
+    # Enforce maximum number of files/images per request
+    if len(files) > MAX_IMAGE_MESSAGES:
+        gr.Warning(f"Too many images provided; keeping only the first {MAX_IMAGE_MESSAGES} file(s).")
+        files = files[:MAX_IMAGE_MESSAGES]
+
     try:
         # Check if the message is empty
         if not message.strip() and len(files) == 0:
@@ -343,6 +351,24 @@ def run_chat_inference(history, message, state):
 
             log_debug(f"sending api_messages to model {model_name}: {api_messages}")
 
+            # Ensure we don't send too many images (count only messages whose content is a list of parts)
+            image_msg_indices = [
+                i for i, msg in enumerate(api_messages)
+                if isinstance(msg, dict) and isinstance(msg.get('content'), list)
+            ]
+            image_count = len(image_msg_indices)
+            if image_count > MAX_IMAGE_MESSAGES:
+                # Remove oldest image messages until we have MAX_IMAGE_MESSAGES or fewer
+                to_remove = image_count - MAX_IMAGE_MESSAGES
+                removed = 0
+                for idx in image_msg_indices:
+                    if removed >= to_remove:
+                        break
+                    # Pop considering prior removals shift indices
+                    api_messages.pop(idx - removed)
+                    removed += 1
+                gr.Warning(f"Too many images provided; keeping the latest {MAX_IMAGE_MESSAGES} and dropped {removed} older image message(s).")
+
             stream = openai_client.chat.completions.create(
                 model=model_name,
                 messages=api_messages,
@@ -598,5 +624,5 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
         api_name=False
     )
 
-demo.queue(default_concurrency_limit=2).launch(ssr_mode=False, show_api=False)
+demo.queue(default_concurrency_limit=2).launch(ssr_mode=False, show_api=False, max_file_size="10mb")
 log_info("Gradio app launched")