Update app.py

app.py

@@ -29,7 +29,7 @@ print("Loading STT model...")
 stt_processor = AutoProcessor.from_pretrained(STT_MODEL_ID)
 stt_model = AutoModelForImageTextToText.from_pretrained(
     STT_MODEL_ID,
-    torch_dtype=DTYPE,
+    torch_dtype="auto",
     device_map="auto",
 )
 
@@ -37,8 +37,8 @@ print("Loading TTS model...")
 tts_tokenizer = AutoTokenizer.from_pretrained(TTS_MODEL_ID)
 tts_model = AutoModelForCausalLM.from_pretrained(
     TTS_MODEL_ID,
-    torch_dtype=DTYPE,
-).to(DEVICE)
+    torch_dtype="auto",
+)
 
 # -----------------------------
 # PIPELINE FUNCTION
@@ -73,7 +73,7 @@ def speech_to_speech(audio_file):
     tts_inputs = tts_tokenizer(
         transcription,
         return_tensors="pt",
-    ).to(DEVICE)
+    )
 
     with torch.no_grad():
         speech = tts_model.generate(**tts_inputs)