Update app.py

app.py

@@ -26,11 +26,11 @@ text_model = AutoModel.from_pretrained('openbmb/MiniCPM-V-2', trust_remote_code=
                                        device_map="auto", torch_dtype=torch.bfloat16)
 tokenizer = AutoTokenizer.from_pretrained('openbmb/MiniCPM-V-2', trust_remote_code=True)
 
-tts_model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-large-v1").to('cuda')
+tts_model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-large-v1")
 tts_tokenizer = AutoTokenizer.from_pretrained("parler-tts/parler-tts-large-v1")
 
-image_model = UNet2DConditionModel.from_config("stabilityai/stable-diffusion-xl-base-1.0", subfolder="unet").to("cuda", torch.float16)
-image_pipe = StableDiffusionXLPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", unet=image_model, torch_dtype=torch.float16, variant="fp16").to("cuda")
+image_model = UNet2DConditionModel.from_config("stabilityai/stable-diffusion-xl-base-1.0", subfolder="unet")
+image_pipe = StableDiffusionXLPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", unet=image_model, torch_dtype=torch.float16, variant="fp16")
 image_pipe.scheduler = EulerDiscreteScheduler.from_config(image_pipe.scheduler.config, timestep_spacing="trailing")
 
 # Initialize voice-only mode
@@ -177,6 +177,9 @@ def initialize_tools():
 # Gradio Interface
 def main_interface(user_prompt, image=None, video=None, audio=None, doc=None, voice_only=False):
     text_model = text_model.to(device='cuda', dtype=torch.bfloat16)
+    tts_model.to("cuda")
+    image_model.to("cuda", torch.float16)
+    image_pip.to("cuda")
     response = handle_input(user_prompt, image=image, video=video, audio=audio, doc=doc)
     if voice_only:
         audio_file = play_voice_output(response)