Update app.py

app.py

@@ -53,6 +53,8 @@ def stream_completion(message: str,
     # Current user input comes last
     messages.append({"role": "user", "content": message})
 
+    os.environ["MIXINPUTS_BETA"] = str(beta)
+
     #try:
     # Kick off streaming completion
     response = openai.chat.completions.create(
@@ -62,8 +64,6 @@ def stream_completion(message: str,
         top_p=top_p,
         max_tokens=max_tokens,
         stream=True,
-        # Forward MoI blending coefficient to the backend
-        extra_headers={"X-MIXINPUTS-BETA": str(beta)},
     )
 
     assistant = ""