changes to cuda structure

generation_utilities.py

@@ -15,7 +15,7 @@ audio_diffusion_v1 = AudioDiffusionPipeline.from_pretrained("SAint7579/orpheus_l
 ddim = AudioDiffusionPipeline.from_pretrained("teticio/audio-diffusion-ddim-256").to(device)
 
 ### Add numpy docstring to generate_from_music
-def generate_from_music(song_array, diffuser, start_step, total_steps=100, device="cuda"):
+def generate_from_music(song_array, diffuser, start_step, total_steps=100, device=device):
     """
     Generates audio from a given song array using a given diffuser.
     Parameters
@@ -40,7 +40,7 @@ def generate_from_music(song_array, diffuser, start_step, total_steps=100, devic
     output = diffuser(raw_audio=song_array, generator = generator, start_step=start_step, steps=total_steps)
     return output.images[0], output.audios[0, 0]
 
-def generate_from_music_long(song_array, diffuser, start_step, total_steps=100, device="cuda"):
+def generate_from_music_long(song_array, diffuser, start_step, total_steps=100, device=device):
     """
     Generates a 10 second audio from a given song array using a given diffuser.
     Parameters
@@ -108,7 +108,7 @@ def iterative_slerp(song_arrays, ddim, steps=10):
 
     return slerp
 
-def merge_songs(song_arrays, ddim, slerp_steps=10, diffusion_steps=100, device="cuda"):
+def merge_songs(song_arrays, ddim, slerp_steps=10, diffusion_steps=100, device=device):
     """Merge songs.
 
     Parameters
@@ -132,7 +132,7 @@ def merge_songs(song_arrays, ddim, slerp_steps=10, diffusion_steps=100, device="
     return merged.images[0], merged.audios[0, 0]
 
 ## Write generate songs function with numpy docstring
-def generate_songs(conditioning_songs, similarity=0.9, quality=500, merging_quality=100, device='cuda'):
+def generate_songs(conditioning_songs, similarity=0.9, quality=500, merging_quality=100, device=device):
     """Generate songs.
 
     Parameters

ui/app.py

@@ -38,7 +38,7 @@ if submit:
     song_list = [librosa.load(os.path.join(os.getcwd(),f"input_songs/{song}.mp3"), sr=22050)[0] for song in song_options]
     print(song_options,print(song_list))
     
-    spectrogram, generated_song, model_name = generation_utilities.generate_songs(song_list, similarity=similarity, quality=500, merging_quality=100, device='cuda')
+    spectrogram, generated_song, model_name = generation_utilities.generate_songs(song_list, similarity=similarity, quality=500, merging_quality=100)
     st.session_state['song_name'] = song_options[0]
     st.session_state['song_list'] = [os.path.join(os.getcwd(),f"input_songs/{song}.mp3") for song in song_options]
     st.session_state['song_name'] = '_'.join(song_options)