Fix spaces bug

modules/whisper/faster_whisper_inference.py

@@ -13,8 +13,6 @@ from argparse import Namespace
 from modules.whisper.whisper_parameter import *
 from modules.whisper.whisper_base import WhisperBase
 
-# ZeroGPU
-import spaces
 
 class FasterWhisperInference(WhisperBase):
     def __init__(self,
@@ -33,7 +31,6 @@ class FasterWhisperInference(WhisperBase):
         self.available_compute_types = self.get_available_compute_type()
         self.download_model(model_size="large-v2", model_dir=self.model_dir)
 
-    @spaces.GPU(duration=120)
     def transcribe(self,
                    audio: Union[str, BinaryIO, np.ndarray],
                    progress: gr.Progress,
@@ -93,7 +90,6 @@ class FasterWhisperInference(WhisperBase):
         print("transcribe: finished")
         return segments_result, elapsed_time
 
-    @spaces.GPU(duration=120)
     def update_model(self,
                      model_size: str,
                      compute_type: str,
@@ -124,40 +120,6 @@ class FasterWhisperInference(WhisperBase):
         )
         print("update_model: finished")
 
-    # debug
-    @spaces.GPU(duration=120)
-    def transcribe_file(self,
-                        files: list,
-                        file_format: str,
-                        add_timestamp: bool,
-                        progress=gr.Progress(),
-                        # *whisper_params,
-                        ) -> list:
-        """
-        Write subtitle file from Files
-
-        Parameters
-        ----------
-        files: list
-            List of files to transcribe from gr.Files()
-        file_format: str
-            Subtitle File format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
-        add_timestamp: bool
-            Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the subtitle filename.
-        progress: gr.Progress
-            Indicator to show progress directly in gradio.
-        *whisper_params: tuple
-            Parameters related with whisper. This will be dealt with "WhisperParameters" data class
-
-        Returns
-        ----------
-        result_str:
-            Result of transcription to return to gr.Textbox()
-        result_file_path:
-            Output file path to return to gr.Files()
-        """
-        print('Transcription START')
-
     def get_model_paths(self):
         """
         Get available models from models path including fine-tuned model.
@@ -188,18 +150,10 @@ class FasterWhisperInference(WhisperBase):
             return ['float32', 'int8_float16', 'float16', 'int8', 'int8_float32']
         return ['int16', 'float32', 'int8', 'int8_float32']
 
-    @staticmethod
-    @spaces.GPU
-    def get_device():
-        print("GET DEVICE:")
-        if torch.cuda.is_available():
-            print("GET DEVICE: device is cuda")
+    def get_device(self):
+        if self.device == "cuda":
             return "cuda"
-        elif torch.backends.mps.is_available():
-            return "auto"
-        else:
-            print("GET DEVICE: device is cpu")
-            return "cpu"
+        return "cpu"
 
     @staticmethod
     def download_model(model_size: str, model_dir: str):

modules/whisper/whisper_base.py

@@ -42,7 +42,6 @@ class WhisperBase(ABC):
         self.vad = SileroVAD()
 
     @abstractmethod
-    @spaces.GPU(duration=120)
     def transcribe(self,
                    audio: Union[str, BinaryIO, np.ndarray],
                    progress: gr.Progress,
@@ -51,7 +50,6 @@ class WhisperBase(ABC):
         pass
 
     @abstractmethod
-    @spaces.GPU(duration=120)
     def update_model(self,
                      model_size: str,
                      compute_type: str,
@@ -59,7 +57,6 @@ class WhisperBase(ABC):
                      ):
         pass
 
-    @spaces.GPU(duration=120)
     def run(self,
             audio: Union[str, BinaryIO, np.ndarray],
             progress: gr.Progress,
@@ -125,43 +122,8 @@ class WhisperBase(ABC):
             elapsed_time += elapsed_time_diarization
         return result, elapsed_time
 
-    #debug
     @spaces.GPU(duration=120)
     def transcribe_file(self,
-                        files: list,
-                        file_format: str,
-                        add_timestamp: bool,
-                        progress=gr.Progress(),
-                        #*whisper_params,
-                        ) -> list:
-        """
-        Write subtitle file from Files
-
-        Parameters
-        ----------
-        files: list
-            List of files to transcribe from gr.Files()
-        file_format: str
-            Subtitle File format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
-        add_timestamp: bool
-            Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the subtitle filename.
-        progress: gr.Progress
-            Indicator to show progress directly in gradio.
-        *whisper_params: tuple
-            Parameters related with whisper. This will be dealt with "WhisperParameters" data class
-
-        Returns
-        ----------
-        result_str:
-            Result of transcription to return to gr.Textbox()
-        result_file_path:
-            Output file path to return to gr.Files()
-        """
-        print('Transcription START')
-
-
-    @spaces.GPU(duration=120)
-    def transcribe_file_releas(self,
                         files: list,
                         file_format: str,
                         add_timestamp: bool,
@@ -438,8 +400,12 @@ class WhisperBase(ABC):
 
         return time_str.strip()
 
+    def release_cuda_memory(self):
+        if self.device == "cuda":
+            torch.cuda.empty_cache()
+            torch.cuda.reset_max_memory_allocated()
+
     @staticmethod
-    @spaces.GPU(duration=120)
     def get_device():
         if torch.cuda.is_available():
             return "cuda"
@@ -448,13 +414,6 @@ class WhisperBase(ABC):
         else:
             return "cpu"
 
-    @staticmethod
-    @spaces.GPU(duration=120)
-    def release_cuda_memory():
-        if torch.cuda.is_available():
-            torch.cuda.empty_cache()
-            torch.cuda.reset_max_memory_allocated()
-
     @staticmethod
     def remove_input_files(file_paths: List[str]):
         if not file_paths: