run vad on gpu

faster_whisper/vad.py

@@ -251,13 +251,13 @@ def __init__(self, path):
             ) from e
 
         opts = onnxruntime.SessionOptions()
-        opts.inter_op_num_threads = 1
-        opts.intra_op_num_threads = 1
         opts.log_severity_level = 4
+        opts.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_BASIC
+        # https://github.com/microsoft/onnxruntime/issues/11548#issuecomment-1158314424
 
         self.session = onnxruntime.InferenceSession(
             path,
-            providers=["CPUExecutionProvider"],
+            providers=["CUDAExecutionProvider"],
             sess_options=opts,
         )
 

requirements.txt

@@ -2,4 +2,4 @@ av==10.*
 ctranslate2>=3.17,<4
 huggingface_hub>=0.13
 tokenizers>=0.13,<0.15
-onnxruntime>=1.14,<2
+onnxruntime-gpu>=1.16.0,<2