HTTP Error 403: rate limit exceeded when loading model

[ichamatidis]

🐛 Bug

To Reproduce

Steps to reproduce the behavior:

1. import torch
2. model = torch.hub.load('pytorch/vision:v1.9.0','resnet50', pretrained=True)

Expected behavior

Load the model

Environment

Google Colab and also local with the latest pytorch and torchvision installed

Error Output

HTTPError Traceback (most recent call last)

in ()
1 import torchvision
2
----> 3 model = torch.hub.load('pytorch/vision:v1.9.0','resnet50', pretrained=True)

8 frames

/usr/lib/python3.7/urllib/request.py in http_error_default(self, req, fp, code, msg, hdrs)
647 class HTTPDefaultErrorHandler(BaseHandler):
648 def http_error_default(self, req, fp, code, msg, hdrs):
--> 649 raise HTTPError(req.full_url, code, msg, hdrs, fp)
650
651 class HTTPRedirectHandler(BaseHandler):

HTTPError: HTTP Error 403: rate limit exceeded

[oke-aditya]

I'm able to download on my local machine. With the following command

import torch
model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet50', pretrained=True)

Also with

import torch
model = torch.hub.load('pytorch/vision:v0.9.0', 'resnet50', pretrained=True)

Can you give one more try @ichamatidis ?

[ichamatidis]

I tried it with both versions that you mentioned, still the same result in Google Colab and locally. I provide bellow the error output form the 0.10.0 version from Google Colab.

Error Output

---

HTTPError Traceback (most recent call last)

in ()
1 import torch
2
----> 3 model = torch.hub.load('pytorch/vision:v0.10.0','resnet50', pretrained=True)

8 frames

/usr/lib/python3.7/urllib/request.py in http_error_default(self, req, fp, code, msg, hdrs)
647 class HTTPDefaultErrorHandler(BaseHandler):
648 def http_error_default(self, req, fp, code, msg, hdrs):
--> 649 raise HTTPError(req.full_url, code, msg, hdrs, fp)
650
651 class HTTPRedirectHandler(BaseHandler):

HTTPError: HTTP Error 403: rate limit exceeded

[oke-aditya]

True. I can reproduce this over Google Colab. Not locally though

Here is link

Unsure why colab should do that.

But you can access pretrained model from
rnet = torchvision.models.resnet50(pretrained=True)

[ichamatidis]

Yeah it's a weird bug and must be recent(less than 2 weeks cause it was working normally before). Does your local machine has Windows or Linux?

[oke-aditya]

Linux Ubuntu 20.04.2

[ichamatidis]

I have Ubuntu 21.04, so I assume it's some connection error. Thank you for your advise :)

[ptrblck]

Issue is also observed here and I'm also able to reproduce it using Ubuntu20.04.

[pmeier]

This seems to be a bug in torch.hub with pytorch/pytorch#56138 as possible offender. I can't reproduce the issue with torch==1.8.0 which does not contain this commit. cc @cyyever @ailzhang

[ailzhang]

Yea I think it's introduced in that PR cc: @cyyever. It's probably hitting the rate limit set by github. My guess is it's only reproducible on CI machines/Google Colab since those machines might send a lot of API requests from the same public IP address so github only errors out on those. Our personal dev machines didn't hit the limit so it's not reproducible there.
If we still don't want to introduce dependency on system git, we probably need to either find another way to validate branch/tag or somehow workaround the rate limit.

[AK391]

torch==1.8.0

can confirm using torch==1.8.0 and restarting runtime in colab works

[Vernerrf]

Hi everyone,
With me the error happens with any model used. And even resetting the collab the problem persists.

[pmeier]

@ailzhang I can reproduce this locally and @NicolasHug seems to have hit the same in #4157. So it is not only CI machines and Colab.

---

Bumping priority due to user activity.