Update app_v1v2.py

app_v1v2.py

@@ -24,11 +24,11 @@ if hasattr(torch._inductor.config, "fx_graph_cache"):
 dtype = torch.float32
 
 # Global variables to store model instances
-vc_wrapper_v1 = None
-vc_wrapper_v2 = None
+vc_wrapper_v1 = SeedVCWrapper()
+vc_wrapper_v2 = load_v2_models(args)
 
 
-def load_v2_models(args):
+def load_v2_models():
     from hydra.utils import instantiate
     from omegaconf import DictConfig
     cfg = DictConfig(yaml.safe_load(open("configs/v2/vc_wrapper.yaml", "r")))
@@ -39,10 +39,6 @@ def load_v2_models(args):
 
     vc_wrapper.setup_ar_caches(max_batch_size=1, max_seq_len=4096, dtype=dtype, device=device)
 
-    if args.compile:
-        vc_wrapper.compile_ar()
-        # vc_wrapper.compile_cfm()
-
     return vc_wrapper
 
 @spaces.GPU
@@ -52,9 +48,6 @@ def convert_voice_v1_wrapper(source_audio_path, target_audio_path, diffusion_ste
     """
     Wrapper function for vc_wrapper.convert_voice that can be decorated with @spaces.GPU
     """
-    global vc_wrapper_v1
-    if vc_wrapper_v1 is None:
-        vc_wrapper_v1 = SeedVCWrapper()
 
     # Use yield from to properly handle the generator
     yield from vc_wrapper_v1.convert_voice(
@@ -77,13 +70,6 @@ def convert_voice_v2_wrapper(source_audio_path, target_audio_path, diffusion_ste
     """
     Wrapper function for vc_wrapper.convert_voice_with_streaming that can be decorated with @spaces.GPU
     """
-    global vc_wrapper_v2
-    if vc_wrapper_v2 is None:
-        # Initialize with default arguments
-        parser = argparse.ArgumentParser()
-        parser.add_argument("--compile", type=bool, default=True)
-        args = parser.parse_args([])
-        vc_wrapper_v2 = load_v2_models(args)
 
     # Use yield from to properly handle the generator
     yield from vc_wrapper_v2.convert_voice_with_streaming(