Error when downloaded via pipeline
I'm trying to use the example code and getting this errors
ChunkedEncodingError: ("Connection broken: ConnectionResetError(104, 'Connection reset by peer')", ConnectionResetError(104, 'Connection reset by peer'))
me too i tried 20 times its stop here
raise ChunkedEncodingError(e)
requests.exceptions.ChunkedEncodingError: ("Connection broken: ConnectionResetError(104, 'Connection reset by peer')", ConnectionResetError(104, 'Connection reset by peer'))
Downloading: 16%|█▋ | 791M/4.86G [12:12<1:04:34, 1.13MB/s]
Downloading from damo/Image-to-Video results in the first error and downloading from here results in the second error:
requests.exceptions.ChunkedEncodingError: ("Connection broken: ConnectionResetError(10054)")
raise ValueError(f'invalid model repo path {e}')
ValueError: invalid model repo path Response details: {'Code': 10010205001, 'Message': 'ÞÄÀÕÅûµ¿íÕ×ïõ┐íµü»Õñ▒Þ┤Ñ´╝îõ┐íµü»´╝Ürecord not found', 'RequestId': 'fa92bb3b-07c4-4337-b7c2-13f1fdf52337', 'Success': False}
I have had users report the downloads always take a few attempts to get all the models. Some take more tries than others.
The models should be moved to a more reliable server (like on huggingface) to host the model downloads.
Looks like the modelscope servers are either too slow, overloaded, or just restricted to give poor downloads.
I apologize for the inconvenience caused. Could you please try again in a different network environment?
It is not just me. I use your scripts in Visions of Chaos.
Many of my users have complained about timeouts and slow downloads. Needing to try multiple times before the models all finally download. It must be an issue with the modelscope servers.
Modelscope also seems to block VPN, so VPN services need to be disabled before downloading.
If you did use Hugging face the downloads would be much faster and more reliable. Many other AI systems use hugging face to host their models.
Other systems I see that use modelscope also have timeout issues reported when downloading the models.
Just a suggestion to improve the user experience :)
THis code fixed it
from modelscope.pipelines import pipeline
from modelscope.outputs import OutputKeys
from modelscope import snapshot_download
model_dir = snapshot_download('damo/Image-to-Video', revision='v1.1.0')
pipe = pipeline(task='image-to-video', model= model_dir, model_revision='v1.1.0')
IMG_PATH = "./test.png"
output_video_path = pipe(IMG_PATH, output_video='./output.mp4')[OutputKeys.OUTPUT_VIDEO]
print(output_video_path)
How does that fix it? Won't the models be downloaded from the same modelscope servers?
Note, the problem is not the models never download (like a bad URL), the problem is they fail during the downloads.
How does that fix it? Won't the models be downloaded from the same modelscope servers?
Note, the problem is not the models never download (like a bad URL), the problem is they fail during the downloads.
No idea but it works very consistently for me now. Both on my PC and quite a few times on RunPod where I get a random networks from all over the world.
Here I even made a post about this on my Blog https://followfoxai.substack.com/p/testing-the-brand-new-chinese-image
In the last few days there have been a couple of fixes and download improvement patches committed, so hopefully will the 0.19 release be better. Here's my report on GitHub: https://github.com/modelscope/modelscope/issues/518