update

app.py

@@ -32,7 +32,8 @@ from src.audio2vid import get_headpose_temp, smooth_pose_seq
 from src.utils.frame_interpolation import init_frame_interpolation_model, batch_images_interpolation_tool
 
 if torch.backends.mps.is_available():
-  device = "mps"
+  #device = "mps"
+  device = "cpu"
 elif torch.cuda.is_available():
   device = "cuda"
 else:
@@ -44,6 +45,9 @@ if config.weight_dtype == "fp16":
     weight_dtype = torch.float16
 else:
     weight_dtype = torch.float32
+
+if device == "cpu":
+    weight_dtype = torch.float32
     
 audio_infer_config = OmegaConf.load(config.audio_inference_config)
 # prepare model
@@ -378,7 +382,10 @@ with gr.Blocks() as demo:
                     a2v_headpose_video = gr.Video(label="Option: upload head pose reference video", sources="upload")
 
                 with gr.Row():
-                    a2v_size_slider = gr.Slider(minimum=256, maximum=1024, step=8, value=512, label="Video size (-W & -H)")
+                    if device == "cpu":
+                        a2v_size_slider = gr.Slider(minimum=256, maximum=1024, step=8, value=256, label="Video size (-W & -H)")
+                    else:
+                        a2v_size_slider = gr.Slider(minimum=256, maximum=1024, step=8, value=512, label="Video size (-W & -H)")
                     a2v_step_slider = gr.Slider(minimum=5, maximum=50, step=1, value=20, label="Steps (--steps)")
                 
                 with gr.Row():
@@ -406,7 +413,10 @@ with gr.Blocks() as demo:
                     v2v_source_video = gr.Video(label="Upload source video", sources="upload")
                 
                 with gr.Row():
-                    v2v_size_slider = gr.Slider(minimum=256, maximum=1024, step=8, value=512, label="Video size (-W & -H)")
+                    if device == "cpu":
+                        v2v_size_slider = gr.Slider(minimum=256, maximum=1024, step=8, value=256, label="Video size (-W & -H)")
+                    else:
+                        v2v_size_slider = gr.Slider(minimum=256, maximum=1024, step=8, value=512, label="Video size (-W & -H)")
                     v2v_step_slider = gr.Slider(minimum=5, maximum=50, step=1, value=20, label="Steps (--steps)")
                 
                 with gr.Row():

src/utils/frame_interpolation.py

@@ -6,7 +6,8 @@ import bisect
 import shutil
 
 if torch.backends.mps.is_available():
-  device = "mps"
+  #device = "mps"
+  device = "cpu"
 elif torch.cuda.is_available():
   device = "cuda"
 else:

src/utils/util.py

@@ -13,7 +13,8 @@ from einops import rearrange
 from PIL import Image
 
 if torch.backends.mps.is_available():
-  device = "mps"
+  device = "cpu"
+  #device = "mps"
 elif torch.cuda.is_available():
   device = "cuda"
 else: