update gradio

app.py

@@ -1,56 +1,74 @@
 import gradio as gr
 import requests
+import sseclient
 import os
-import spaces
-
-from server import setup_mixinputs, launch_vllm_server
 
 API_URL = "http://localhost:8000/v1/chat/completions"
 
-def chat_with_moi(message, history, temperature, top_p, beta):
-    # Set the MIXINPUTS_BETA env var *per request*
-    os.environ["MIXINPUTS_BETA"] = str(beta)
-
-    # setup_mixinputs()
-    # launch_vllm_server(beta=beta)
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+    beta,
+):
+    # Build message history
+    messages = [{"role": "system", "content": system_message}]
+    for user, assistant in history:
+        if user:
+            messages.append({"role": "user", "content": user})
+        if assistant:
+            messages.append({"role": "assistant", "content": assistant})
+    messages.append({"role": "user", "content": message})
 
+    # Prepare request payload
     payload = {
-        "model": "Qwen/Qwen3-4B",  # match what your vLLM server expects
-        "messages": [{"role": "user", "content": message}],
+        "model": "Qwen/Qwen3-4B",  # Update to your actual model if needed
+        "messages": messages,
         "temperature": temperature,
         "top_p": top_p,
-        "max_tokens": 512,
+        "max_tokens": max_tokens,
+        "stream": True,
     }
 
+    # Optional: send beta as a custom OpenAI field
+    headers = {
+        "Content-Type": "application/json",
+        "X-MIXINPUTS-BETA": str(beta),  # or modify your vLLM code to read this
+    }
+
+    # Stream response using SSE (Server-Sent Events)
     try:
-        response = requests.post(API_URL, json=payload)
+        response = requests.post(API_URL, json=payload, stream=True, headers=headers)
         response.raise_for_status()
-        return response.json()["choices"][0]["message"]["content"]
-    except Exception as e:
-        return f"[ERROR] {str(e)}"
+        client = sseclient.SSEClient(response)
 
-# Gradio UI
-with gr.Blocks() as demo:
-    gr.Markdown("# 🧪 Mixture of Inputs (MoI) Demo with vLLM")
+        full_text = ""
+        for event in client.events():
+            if event.data == "[DONE]":
+                break
+            delta = event.json()["choices"][0]["delta"].get("content", "")
+            full_text += delta
+            yield full_text
 
-    with gr.Row():
-        temperature = gr.Slider(0.0, 1.5, value=0.7, label="Temperature")
-        top_p = gr.Slider(0.0, 1.0, value=0.95, label="Top-p")
-        beta = gr.Slider(0.0, 10.0, value=1.0, label="MoI Beta")
-
-    chatbot = gr.Chatbot(type="messages")
-    message = gr.Textbox(label="Your message")
-    send_btn = gr.Button("Send")
-
-    history = gr.State([])
-
-    def respond(user_message, chat_history, temperature, top_p, beta):
-        reply = chat_with_moi(user_message, chat_history, temperature, top_p, beta)
-        chat_history = chat_history + [{"role": "user", "content": user_message},
-                                    {"role": "assistant", "content": reply}]
-        return chat_history, chat_history
+    except Exception as e:
+        yield f"[ERROR] {e}"
 
-    send_btn.click(respond, inputs=[message, history, temperature, top_p, beta],
-                   outputs=[chatbot, history])
+# UI layout using ChatInterface
+demo = gr.ChatInterface(
+    respond,
+    additional_inputs=[
+        gr.Textbox(value="You are a helpful assistant using Mixture of Inputs.", label="System message"),
+        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
+        gr.Slider(minimum=0.0, maximum=10.0, value=1.0, step=0.1, label="MoI Beta"),
+    ],
+    title="🧪 Mixture of Inputs (MoI) Demo",
+    description="Streaming local vLLM demo with dynamic MoI beta adjustment.",
+)
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()