Supports Intel GPU

backend/configs/config.yaml

@@ -10,7 +10,7 @@ bgm_separation:
   # Whether to offload the model after the inference. Should be true if your setup has a VRAM less than <16GB
   enable_offload: true
   # Device to load BGM separation model
-  device: cuda
+  device: xpu
 
 # Settings that apply to the `cache' directory. The output files for `/bgm-separation` are stored in the `cache' directory,
 # (You can check out the actual generated files by testing `/bgm-separation`.)

backend/tests/test_backend_bgm_separation.py

@@ -12,7 +12,7 @@
 )
 
 
-@pytest.mark.skipif(not torch.cuda.is_available(), reason="Skip the test because CUDA is not available")
+@pytest.mark.skipif(not torch.xpu.is_available(), reason="Skip the test because CUDA is not available")
 @pytest.mark.parametrize(
     "bgm_separation_params",
     [

configs/default_parameters.yaml

@@ -1,64 +1,63 @@
 whisper:
-  model_size: "large-v2"
-  file_format: "SRT"
-  lang: "Automatic Detection"
+  model_size: large-v2
+  lang: chinese
   is_translate: false
   beam_size: 5
-  log_prob_threshold: -1
+  log_prob_threshold: -1.0
   no_speech_threshold: 0.6
+  compute_type: float32
   best_of: 5
-  patience: 1
+  patience: 1.0
   condition_on_previous_text: true
   prompt_reset_on_temperature: 0.5
   initial_prompt: null
-  temperature: 0
+  temperature: 0.0
   compression_ratio_threshold: 2.4
-  chunk_length: 30
-  batch_size: 24
-  length_penalty: 1
-  repetition_penalty: 1
+  length_penalty: 1.0
+  repetition_penalty: 1.0
   no_repeat_ngram_size: 0
   prefix: null
   suppress_blank: true
-  suppress_tokens: "[-1]"
-  max_initial_timestamp: 1
+  suppress_tokens: '[-1]'
+  max_initial_timestamp: 1.0
   word_timestamps: false
-  prepend_punctuations: "\"'“¿([{-"
-  append_punctuations: "\"'.。,，!！?？:：”)]}、"
+  prepend_punctuations: '"''“¿([{-'
+  append_punctuations: '"''.。,，!！?？:：”)]}、'
   max_new_tokens: null
+  chunk_length: 30
   hallucination_silence_threshold: null
   hotwords: null
   language_detection_threshold: 0.5
   language_detection_segments: 1
+  batch_size: 24
   add_timestamp: true
-
+  file_format: SRT
 vad:
   vad_filter: false
   threshold: 0.5
   min_speech_duration_ms: 250
   max_speech_duration_s: 9999
   min_silence_duration_ms: 1000
   speech_pad_ms: 2000
-
 diarization:
   is_diarize: false
-  hf_token: ""
-
+  diarization_device: xpu
+  hf_token: ''
 bgm_separation:
   is_separate_bgm: false
-  uvr_model_size: "UVR-MDX-NET-Inst_HQ_4"
+  uvr_model_size: UVR-MDX-NET-Inst_HQ_4
+  uvr_device: xpu
   segment_size: 256
   save_file: false
   enable_offload: true
-
 translation:
   deepl:
-    api_key: ""
+    api_key: ''
     is_pro: false
-    source_lang: "Automatic Detection"
-    target_lang: "English"
+    source_lang: Automatic Detection
+    target_lang: English
   nllb:
-    model_size: "facebook/nllb-200-1.3B"
+    model_size: facebook/nllb-200-1.3B
     source_lang: null
     target_lang: null
     max_length: 200

modules/diarize/diarizer.py

@@ -129,15 +129,15 @@ def offload(self):
         if self.pipe is not None:
             del self.pipe
             self.pipe = None
-        if self.device == "cuda":
-            torch.cuda.empty_cache()
-            torch.cuda.reset_max_memory_allocated()
+        if self.device == "xpu":
+            torch.xpu.empty_cache()
+            torch.xpu.reset_max_memory_allocated()
         gc.collect()
 
     @staticmethod
     def get_device():
-        if torch.cuda.is_available():
-            return "cuda"
+        if torch.xpu.is_available():
+            return "xpu"
         elif torch.backends.mps.is_available():
             return "mps"
         else:
@@ -146,8 +146,8 @@ def get_device():
     @staticmethod
     def get_available_device():
         devices = ["cpu"]
-        if torch.cuda.is_available():
-            devices.append("cuda")
+        if torch.xpu.is_available():
+            devices.append("xpu")
         elif torch.backends.mps.is_available():
             devices.append("mps")
         return devices

modules/translation/translation_base.py

@@ -127,31 +127,31 @@ def translate_file(self,
             print(f"Error translating file: {e}")
             raise
         finally:
-            self.release_cuda_memory()
+            self.release_xpu_memory()
 
     def offload(self):
         """Offload the model and free up the memory"""
         if self.model is not None:
             del self.model
             self.model = None
-        if self.device == "cuda":
-            self.release_cuda_memory()
+        if self.device == "xpu":
+            self.release_xpu_memory()
         gc.collect()
 
     @staticmethod
     def get_device():
-        if torch.cuda.is_available():
-            return "cuda"
+        if torch.xpu.is_available():
+            return "xpu"
         elif torch.backends.mps.is_available():
             return "mps"
         else:
             return "cpu"
 
     @staticmethod
-    def release_cuda_memory():
-        if torch.cuda.is_available():
-            torch.cuda.empty_cache()
-            torch.cuda.reset_max_memory_allocated()
+    def release_xpu_memory():
+        if torch.xpu.is_available():
+            torch.xpu.empty_cache()
+            torch.xpu.reset_max_memory_allocated()
 
     @staticmethod
     def remove_input_files(file_paths: List[str]):

modules/uvr/music_separator.py

@@ -20,7 +20,7 @@ def __init__(self,
                  output_dir: Optional[str] = UVR_OUTPUT_DIR):
         self.model = None
         self.device = self.get_device()
-        self.available_devices = ["cpu", "cuda"]
+        self.available_devices = ["cpu", "xpu"]
         self.model_dir = model_dir
         self.output_dir = output_dir
         instrumental_output_dir = os.path.join(self.output_dir, "instrumental")
@@ -159,15 +159,15 @@ def separate_files(self,
     @staticmethod
     def get_device():
         """Get device for the model"""
-        return "cuda" if torch.cuda.is_available() else "cpu"
+        return "xpu" if torch.xpu.is_available() else "cpu"
 
     def offload(self):
         """Offload the model and free up the memory"""
         if self.model is not None:
             del self.model
             self.model = None
-        if self.device == "cuda":
-            torch.cuda.empty_cache()
+        if self.device == "xpu":
+            torch.xpu.empty_cache()
         gc.collect()
         self.audio_info = None
 

modules/whisper/base_transcription_pipeline.py

@@ -265,7 +265,7 @@ def transcribe_file(self,
         except Exception as e:
             raise RuntimeError(f"Error transcribing file: {e}") from e
         finally:
-            self.release_cuda_memory()
+            self.release_xpu_memory()
 
     def transcribe_mic(self,
                        mic_audio: str,
@@ -328,7 +328,7 @@ def transcribe_mic(self,
         except Exception as e:
             raise RuntimeError(f"Error transcribing mic: {e}") from e
         finally:
-            self.release_cuda_memory()
+            self.release_xpu_memory()
 
     def transcribe_youtube(self,
                            youtube_link: str,
@@ -400,7 +400,7 @@ def transcribe_youtube(self,
         except Exception as e:
             raise RuntimeError(f"Error transcribing youtube: {e}") from e
         finally:
-            self.release_cuda_memory()
+            self.release_xpu_memory()
 
     def get_compute_type(self):
         if "float16" in self.available_compute_types:
@@ -421,8 +421,8 @@ def offload(self):
         if self.model is not None:
             del self.model
             self.model = None
-        if self.device == "cuda":
-            self.release_cuda_memory()
+        if self.device == "xpu":
+            self.release_xpu_memory()
         gc.collect()
 
     @staticmethod
@@ -454,8 +454,8 @@ def format_time(elapsed_time: float) -> str:
 
     @staticmethod
     def get_device():
-        if torch.cuda.is_available():
-            return "cuda"
+        if torch.xpu.is_available():
+            return "xpu"
         elif torch.backends.mps.is_available():
             if not BaseTranscriptionPipeline.is_sparse_api_supported():
                 # Device `SparseMPS` is not supported for now. See : https://github.com/pytorch/pytorch/issues/87886
@@ -482,11 +482,11 @@ def is_sparse_api_supported():
             return False
 
     @staticmethod
-    def release_cuda_memory():
+    def release_xpu_memory():
         """Release memory"""
-        if torch.cuda.is_available():
-            torch.cuda.empty_cache()
-            torch.cuda.reset_max_memory_allocated()
+        if torch.xpu.is_available():
+            torch.xpu.empty_cache()
+            torch.xpu.max_memory_allocated()
 
     @staticmethod
     def remove_input_files(file_paths: List[str]):

modules/whisper/data_classes.py

@@ -156,7 +156,7 @@ def to_gradio_inputs(cls, defaults: Optional[Dict] = None) -> List[gr.components
 class DiarizationParams(BaseParams):
     """Speaker diarization parameters"""
     is_diarize: bool = Field(default=False, description="Enable speaker diarization")
-    diarization_device: str = Field(default="cuda", description="Device to run Diarization model.")
+    diarization_device: str = Field(default="xpu", description="Device to run Diarization model.")
     hf_token: str = Field(
         default="",
         description="Hugging Face token for downloading diarization models"
@@ -174,7 +174,7 @@ def to_gradio_inputs(cls,
             ),
             gr.Dropdown(
                 label=_("Device"),
-                choices=["cpu", "cuda"] if available_devices is None else available_devices,
+                choices=["cpu", "xpu"] if available_devices is None else available_devices,
                 value=defaults.get("device", device),
             ),
             gr.Textbox(
@@ -192,7 +192,7 @@ class BGMSeparationParams(BaseParams):
         default="UVR-MDX-NET-Inst_HQ_4",
         description="UVR model size"
     )
-    uvr_device: str = Field(default="cuda", description="Device to run UVR model.")
+    uvr_device: str = Field(default="xpu", description="Device to run UVR model.")
     segment_size: int = Field(
         default=256,
         gt=0,
@@ -228,7 +228,7 @@ def to_gradio_input(cls,
             ),
             gr.Dropdown(
                 label=_("Device"),
-                choices=["cpu", "cuda"] if available_devices is None else available_devices,
+                choices=["cpu", "xpu"] if available_devices is None else available_devices,
                 value=defaults.get("device", device),
             ),
             gr.Number(

modules/whisper/faster_whisper_inference.py

@@ -183,8 +183,8 @@ def get_model_paths(self):
 
     @staticmethod
     def get_device():
-        if torch.cuda.is_available():
-            return "cuda"
+        if torch.xpu.is_available():
+            return "xpu"
         else:
             return "auto"
 

requirements.txt

@@ -5,10 +5,9 @@
 --extra-index-url https://download.pytorch.org/whl/cu124
 
 
-torch
-torchaudio
+
 git+https://github.com/jhj0517/jhj0517-whisper.git
-faster-whisper==1.1.1
+#faster-whisper==1.1.1
 transformers
 gradio
 gradio-i18n

tests/test_bgm_separation.py

@@ -11,7 +11,7 @@
 
 
 @pytest.mark.skipif(
-    not is_cuda_available(),
+    not is_xpu_available(),
     reason="Skipping because the test only works on GPU"
 )
 @pytest.mark.parametrize(
@@ -32,7 +32,7 @@ def test_bgm_separation_pipeline(
 
 
 @pytest.mark.skipif(
-    not is_cuda_available(),
+    not is_xpu_available(),
     reason="Skipping because the test only works on GPU"
 )
 @pytest.mark.parametrize(

tests/test_config.py

@@ -21,8 +21,8 @@
 
 
 @functools.lru_cache
-def is_cuda_available():
-    return torch.cuda.is_available()
+def is_xpu_available():
+    return torch.xpu.is_available()
 
 
 @functools.lru_cache

tests/test_diarization.py

@@ -10,7 +10,7 @@
 
 
 @pytest.mark.skipif(
-    not is_cuda_available(),
+    not is_xpu_available(),
     reason="Skipping because the test only works on GPU"
 )
 @pytest.mark.parametrize(