update

app.py

@@ -1,3 +1,4 @@
+import spaces
 import gradio as gr
 import os
 import numpy as np
@@ -11,9 +12,6 @@ huggingface-cli download stabilityai/stable-video-diffusion-img2vid-xt --local-d
 huggingface-cli download openai/whisper-tiny --local-dir checkpoints/whisper-tiny;'
 os.system(cmd)
 
-# import torch
-# print('torch.cuda.device_count()', torch.cuda.device_count())
-
 pipe = Sonic()
 
 def get_md5(content):
@@ -21,6 +19,7 @@ def get_md5(content):
     md5 = md5hash.hexdigest()
     return md5
 
+@spaces.GPU(duration=120)
 def get_video_res(img_path, audio_path, res_video_path, dynamic_scale=1.0):
 
     expand_ratio = 0.5

config/inference/sonic.yaml

@@ -12,7 +12,7 @@ decode_chunk_size: 8
 motion_bucket_scale: 1.0
 image_size: 512
 area: 1.1
-frame_num: 100
+frame_num: 50
 step: 2
 overlap: 0
 shift_offset: 7

sonic.py

@@ -19,12 +19,10 @@ from src.models.audio_adapter.audio_proj import AudioProjModel
 from src.models.audio_adapter.audio_to_bucket import Audio2bucketModel
 from src.utils.RIFE.RIFE_HDv3 import RIFEModel
 from src.dataset.face_align.align import AlignImage
-import spaces
 
 
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
 
-@torch.no_grad()
 def test(
     pipe,
     config,
@@ -256,7 +254,7 @@ class Sonic():
         crop_image = face_image[crop_bbox[1]:crop_bbox[3], crop_bbox[0]:crop_bbox[2]]
         cv2.imwrite(output_image_path, crop_image)
 
-    @spaces.GPU(duration=120)
+    @torch.no_grad()
     def process(self,
                 image_path,
                 audio_path,