feat: Implement self-contained specialist managers

app.py

@@ -14,6 +14,12 @@
 import spaces
 import subprocess
 
+# Clone the repository to ensure all files are available
+subprocess.run("git lfs install", shell=True, check=True)
+subprocess.run("git clone https://huggingface.co/spaces/ByteDance-Seed/SeedVR2-3B", shell=True, check=True)
+os.chdir('SeedVR2-3B')
+
+
 import os
 import torch
 import mediapy
@@ -77,75 +83,9 @@ subprocess.run(
     shell=True,
 )
 
-def load_file_from_url(url, model_dir=None, progress=True, file_name=None):
-    """Load file from http url, will download models if necessary.
-
-    Reference: https://github.com/1adrianb/face-alignment/blob/master/face_alignment/utils.py
-
-    Args:
-        url (str): URL to be downloaded.
-        model_dir (str): The path to save the downloaded model. Should be a full path. If None, use pytorch hub_dir.
-            Default: None.
-        progress (bool): Whether to show the download progress. Default: True.
-        file_name (str): The downloaded file name. If None, use the file name in the url. Default: None.
-
-    Returns:
-        str: The path to the downloaded file.
-    """
-    if model_dir is None:  # use the pytorch hub_dir
-        hub_dir = get_dir()
-        model_dir = os.path.join(hub_dir, 'checkpoints')
-
-    os.makedirs(model_dir, exist_ok=True)
-
-    parts = urlparse(url)
-    filename = os.path.basename(parts.path)
-    if file_name is not None:
-        filename = file_name
-    cached_file = os.path.abspath(os.path.join(model_dir, filename))
-    if not os.path.exists(cached_file):
-        print(f'Downloading: "{url}" to {cached_file}\n')
-        download_url_to_file(url, cached_file, hash_prefix=None, progress=progress)
-    return cached_file
-
-
-# os.system("pip freeze")
-ckpt_dir = Path('./ckpts')
-if not ckpt_dir.exists():
-	ckpt_dir.mkdir()
-
-pretrain_model_url = {
-	'vae': 'https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/ema_vae.pth',
-	'dit': 'https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/seedvr2_ema_3b.pth',
-    'pos_emb': 'https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/pos_emb.pt',
-    'neg_emb': 'https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/neg_emb.pt',
-    'apex': 'https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/apex-0.1-cp310-cp310-linux_x86_64.whl'
-}
-# download weights
-if not os.path.exists('./ckpts/seedvr2_ema_3b.pth'):
-	load_file_from_url(url=pretrain_model_url['dit'], model_dir='./ckpts/', progress=True, file_name=None)
-if not os.path.exists('./ckpts/ema_vae.pth'):
-	load_file_from_url(url=pretrain_model_url['vae'], model_dir='./ckpts/', progress=True, file_name=None)
-if not os.path.exists('./pos_emb.pt'):
-	load_file_from_url(url=pretrain_model_url['pos_emb'], model_dir='./', progress=True, file_name=None)
-if not os.path.exists('./neg_emb.pt'):
-	load_file_from_url(url=pretrain_model_url['neg_emb'], model_dir='./', progress=True, file_name=None)
-if not os.path.exists('./apex-0.1-cp310-cp310-linux_x86_64.whl'):
-	load_file_from_url(url=pretrain_model_url['apex'], model_dir='./', progress=True, file_name=None)
-
 subprocess.run(shlex.split("pip install apex-0.1-cp310-cp310-linux_x86_64.whl"))
 print(f"✅ setup completed Apex")
 
-# download images
-torch.hub.download_url_to_file(
-	'https://huggingface.co/datasets/Iceclear/SeedVR_VideoDemos/resolve/main/seedvr_videos_crf23/aigc1k/23_1_lq.mp4',
-	'01.mp4')
-torch.hub.download_url_to_file(
-	'https://huggingface.co/datasets/Iceclear/SeedVR_VideoDemos/resolve/main/seedvr_videos_crf23/aigc1k/28_1_lq.mp4',
-	'02.mp4')
-torch.hub.download_url_to_file(
-	'https://huggingface.co/datasets/Iceclear/SeedVR_VideoDemos/resolve/main/seedvr_videos_crf23/aigc1k/2_1_lq.mp4',
-	'03.mp4')
 
 def configure_sequence_parallel(sp_size):
     if sp_size > 1:
@@ -506,4 +446,4 @@ with gr.Blocks(title="SeedVR2: One-Step Video Restoration via Diffusion Adversar
     """)
 
 demo.queue()
-demo.launch()
+demo.launch()