Update gradio_mix.py

gradio_mix.py

@@ -642,7 +642,9 @@ def get_transcribe_state(segments):
         "word_bounds": [f"{word['start']} {word['word']} {word['end']}" for word in segments["words"]]
     }
 
-
+@spaces.GPU
+@torch.no_grad()
+@torch.inference_mode()
 def transcribe(seed, audio_info):
     if transcribe_model is None:
         raise gr.Error("Transcription model not loaded")
@@ -659,6 +661,9 @@ def transcribe(seed, audio_info):
         state
     ]
 
+@spaces.GPU
+@torch.no_grad()
+@torch.inference_mode()
 def align(transcript, audio_info, state):
     lang = state["segments"]["lang"]
     # print("realign: ", transcript, state)
@@ -683,7 +688,9 @@ def align(transcript, audio_info, state):
         state
     ]
 
-
+@spaces.GPU
+@torch.no_grad()
+@torch.inference_mode()
 def denoise(audio_info):
     denoised_audio, sr = denoise_model.denoise(audio_info)
     denoised_audio = denoised_audio # .squeeze().numpy()
@@ -725,7 +732,9 @@ def replace_numbers_with_words(sentence, lang="en"):
             return num # In case num2words fails (unlikely with digits but just to be safe)
     return re.sub(r'\b\d+\b', replace_with_words, sentence) # Regular expression that matches numbers
 
-
+@spaces.GPU
+@torch.no_grad()
+@torch.inference_mode()
 def run(seed, nfe_step, speed, cfg_strength, sway_sampling_coef, ref_ratio,
         audio_info, denoised_audio, transcribe_state, transcript, smart_transcript,
         mode, start_time, end_time,