update

app.py

@@ -1,64 +1,56 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+import requests
+import os
+import spaces
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+from server import setup_mixinputs, launch_vllm_server
 
+API_URL = "http://localhost:8000/v1/chat/completions"
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+@spaces.GPU(duration=120)
+def chat_with_moi(message, history, temperature, top_p, beta):
+    # Set the MIXINPUTS_BETA env var *per request*
+    os.environ["MIXINPUTS_BETA"] = str(beta)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    setup_mixinputs()
+    launch_vllm_server(beta=beta)
 
-    messages.append({"role": "user", "content": message})
+    payload = {
+        "model": "Qwen/QwQ-32B",  # match what your vLLM server expects
+        "messages": [{"role": "user", "content": message}],
+        "temperature": temperature,
+        "top_p": top_p,
+        "max_tokens": 512,
+    }
 
-    response = ""
+    try:
+        response = requests.post(API_URL, json=payload)
+        response.raise_for_status()
+        return response.json()["choices"][0]["message"]["content"]
+    except Exception as e:
+        return f"[ERROR] {str(e)}"
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+# Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("# 🧪 Mixture of Inputs (MoI) Demo with vLLM")
 
-        response += token
-        yield response
+    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()
+    message = gr.Textbox(label="Your message")
+    send_btn = gr.Button("Send")
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", 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)",
-        ),
-    ],
-)
+    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 + [(user_message, reply)]
+        return chat_history, chat_history
 
-if __name__ == "__main__":
-    demo.launch()
+    send_btn.click(respond, inputs=[message, history, temperature, top_p, beta],
+                   outputs=[chatbot, history])
+
+demo.launch()

requirements.txt

@@ -1 +1,3 @@
-huggingface_hub==0.25.2
+huggingface_hub==0.25.2
+vllm==0.8.5
+mixinputs

server.py

@@ -0,0 +1,34 @@
+import subprocess
+import threading
+import os
+import time
+
+def setup_mixinputs():
+    # Step 1: Run mixinputs setup
+    subprocess.run(["mixinputs", "setup"], check=True)
+
+def launch_vllm_server(beta=1.0):
+    # Step 2: Set environment variables
+    env = os.environ.copy()
+    env["MIXINPUTS_BETA"] = str(beta)
+    env["VLLM_USE_V1"] = "1"
+
+    # Step 3: Launch vLLM with custom options
+    cmd = [
+        "vllm", "serve",
+        "Qwen/QwQ-32B",
+        "--tensor-parallel-size", "1",
+        "--enforce-eager",
+        "--max-seq-len-to-capture", "2048",
+        "--max-num-seqs", "1"
+    ]
+    subprocess.run(cmd, env=env)
+
+# Step 1: Setup
+setup_mixinputs()
+
+# Step 2: Launch vLLM server in background
+threading.Thread(target=launch_vllm_server, daemon=True).start()
+
+# Step 3: Give time for server to initialize
+time.sleep(20)