space.gpu

app.py

@@ -4,6 +4,7 @@
 import os
 import yaml
 
+import spaces
 import gradio as gr
 
 import librosa
@@ -22,6 +23,8 @@ if torch.cuda.is_available():
 else:
     device = 'cpu'
 
+
+@spaces.GPU
 def load_laionclap():
     model = laion_clap.CLAP_Module(enable_fusion=True, amodel='HTSAT-tiny').to(device)
     model.load_ckpt(ckpt='630k-audioset-fusion-best.pt')
@@ -88,6 +91,7 @@ def load_audio(file_path, target_sr=44100, duration=33.25, start=0.0):
     return data
 
 
+@spaces.GPU
 @torch.no_grad()
 def compute_laionclap_text_audio_sim(audio_file, laionclap_model, outputs):
     try:
@@ -137,6 +141,7 @@ model = prepare_model(
 )
 
 
+@spaces.GPU
 def inference_item(name, prompt):
     item = {
         'name': str(name),