Fix claude mistake v3

app.py

@@ -2,6 +2,8 @@ import os
 import tempfile
 
 import gradio as gr
+import numpy as np
+import soundfile as sf
 import torch
 
 # --- Device Setup (Explicitly set to CPU) ---
@@ -23,6 +25,7 @@ chatbot_model.to(device)
 from TTS.api import TTS
 
 TTS_MODEL_NAME = "tts_models/en/ljspeech/tacotron2-DDC"
+# Initialize the TTS model on CPU
 tts_model = TTS(model_name=TTS_MODEL_NAME, progress_bar=False)
 
 
@@ -66,10 +69,12 @@ def chat_with_bot(message, history, chat_history_ids=None):
         # Encode the new user input and add end-of-sequence token
         new_input_ids = chatbot_tokenizer.encode(
             message + chatbot_tokenizer.eos_token, return_tensors="pt"
-        )
+        ).to(device)
 
         # Append the new user input tokens to the chat history
         if chat_history_ids is not None:
+            # Move existing chat history IDs to the correct device
+            chat_history_ids = chat_history_ids.to(device)
             bot_input_ids = torch.cat([chat_history_ids, new_input_ids], dim=-1)
         else:
             bot_input_ids = new_input_ids
@@ -109,7 +114,8 @@ custom_css = """
 }
 """
 
-with gr.Blocks(css=custom_css) as demo:
+# CRITICAL FIX: Removed css argument from gr.Blocks()
+with gr.Blocks() as demo:
     gr.Markdown("# 🗣️ STT, TTS & Chat App (CPU Only)")
     gr.Markdown(
         "**NOTE:** This app is running on CPU-only hardware. Speech-to-Text (Whisper) is fast, but **Text-to-Speech (Coqui TTS) and Chat will be slow**."
@@ -185,4 +191,5 @@ with gr.Blocks(css=custom_css) as demo:
                 fn=text_to_speech, inputs=text_input, outputs=[audio_output, tts_status]
             )
 
-demo.launch()
+# CRITICAL FIX: Passed css argument to demo.launch()
+demo.launch(css=custom_css)