add zeroGPU option

hf_gradio_app.py

@@ -1,4 +1,5 @@
 import os, random, time
+import spaces
 import uuid
 import tempfile, shutil
 from pydub import AudioSegment
@@ -78,7 +79,7 @@ def process_audio(file_path, temp_dir):
     audio = AudioSegment.from_file(file_path)
     
     # Check and cut the audio if longer than 4 seconds
-    max_duration = 8 * 1000  # 4 seconds in milliseconds
+    max_duration = 4 * 1000  # 4 seconds in milliseconds
     if len(audio) > max_duration:
         audio = audio[:max_duration]
     
@@ -90,7 +91,8 @@ def process_audio(file_path, temp_dir):
     print(f"Processed audio saved at: {output_path}")
     return output_path
 
-@torch.inference_mode()
+#@torch.inference_mode()
+@spaces.GPU(duration=200)
 def generate(input_video, input_audio, seed, progress=gr.Progress(track_tqdm=True)):
 
     is_shared_ui = True if "fffiloni/MEMO" in os.environ['SPACE_ID'] else False
@@ -194,7 +196,7 @@ def generate(input_video, input_audio, seed, progress=gr.Progress(track_tqdm=Tru
 with gr.Blocks(analytics_enabled=False) as demo:
     with gr.Column():
         gr.Markdown("# MEMO: Memory-Guided Diffusion for Expressive Talking Video Generation")
-        gr.Markdown("Note: On fffiloni's shared UI, audio length is trimmed to max 8 seconds, so everyone can get a taste without to much waiting time in queue.")
+        gr.Markdown("Note: On fffiloni's shared UI, audio length is trimmed to max 4 seconds, so everyone can get a taste without to much waiting time in queue.")
         gr.Markdown("Duplicate the space to skip the queue and enjoy full length capacity.")
         gr.HTML("""
         <div style="display:flex;column-gap:4px;">