Update app.py

app.py

@@ -5,8 +5,9 @@ import numpy as np
 import string
 from huggingface_hub import InferenceClient
 import os
+
 # Initialize Qwen client
-qwen_client = InferenceClient("nztinversive/llama3.2-1b-Uncensored", token=os.environ.get("HF_TOKEN"))
+qwen_client = InferenceClient("nztinversive/llama3.2-1b-Uncensored")
 
 # Model and ONNX setup
 HG_MODEL = "livekit/turn-detector"
@@ -76,7 +77,7 @@ def respond(
     top_p,
 ):
     # Keep the last 4 conversation pairs (user-assistant)
-    messages = [{"role": "system", "content": os.environ.get('CHARACTER_DESC')}]
+    messages = [{"role": "system", "content": os.environ.get("CHARACTER_DESC")}]
 
     for val in history[-10:]:  # Only use the last 4 pairs
         if val[0]: