Update app.py

app.py

@@ -640,6 +640,54 @@ def build_demo():
 
     return demo
 
+# --- API endpoint: /chat ---
+api_interface = gr.Interface(
+    fn=lambda image, question: http_api_infer(image, question),
+    inputs=[gr.Image(type="pil"), gr.Textbox()],
+    outputs="text",
+    allow_flagging="never",
+    api_name="/chat"
+)
+
+def http_api_infer(image, question):
+    """
+    Simple API endpoint that mimics InternVL logic with one image + text.
+    """
+    # Build a simplified version of `state` object here
+    dummy_state = init_state()
+    dummy_state.set_system_message("You are a vision-language assistant.")
+    dummy_state.append_message(Conversation.USER, question, [image])
+    dummy_state.skip_next = False
+
+    # Simulate inference (you can directly call your model function here instead)
+    worker_addr = os.environ.get("WORKER_ADDR", "")
+    api_token = os.environ.get("API_TOKEN", "")
+    headers = {
+        "Authorization": f"Bearer {api_token}", 
+        "Content-Type": "application/json"
+    }
+
+    if not worker_addr:
+        return "⚠️ Model backend is not configured."
+
+    all_image_paths = [dummy_state.save_image(image)]
+
+    pload = {
+        "model": "InternVL2.5-78B",
+        "messages": dummy_state.get_prompt_v2(inlude_image=True, max_dynamic_patch=12),
+        "temperature": 0.2,
+        "top_p": 0.7,
+        "max_tokens": 1024,
+        "repetition_penalty": 1.1,
+        "stream": False
+    }
+
+    try:
+        response = requests.post(worker_addr, json=pload, headers=headers, timeout=120)
+        reply = response.json()["choices"][0]["message"]["content"]
+        return reply
+    except Exception as e:
+        return f"Error: {str(e)}"
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
@@ -652,10 +700,9 @@ if __name__ == "__main__":
     logger.info(f"args: {args}")
 
     logger.info(args)
-    demo = build_demo()
-    demo.queue(api_open=False).launch(
-        server_name=args.host,
-        server_port=args.port,
-        share=args.share,
-        max_threads=args.concurrency_count,
-    )
+    demo = gr.TabbedInterface(
+    interface_list=[build_demo(), api_interface],
+    tab_names=["UI", "API"]
+)
+demo.queue(api_open=True).launch(...)
+