New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fix learning rate history and learning rate computation is schedulers #43
Conversation
Two things can be concluded: * The current computation of the exponential learning rate is incorrect in all version of PyTorch above 0.4.1 * PyTorch 1.4.0 introduced a different design for the LRScheduler; the fix for the exponential learning rate for version 0.4.1 is not the same as the one for 1.4.0 Setup for the above: 1. torch==0.4.1 torchvision==0.2.1 2. torch==1.4.0 torchvision==0.5.0
Also, added caching to make the CI job faster.
Bahhhh, github actions seems to only have PyTorch 1.4.0 and 1.5.0, so we can't run the tests on older versions 😞 . Made a post on the community forum, hopefully I did something wrong or there's a workaround |
Nice! Give me some time to do this! |
I've run this revision on my machine with python-[3.6, 3.7] + torch-[0.4, 1.4], all works fine. 👍 And, good news! For the problem of installing older version of PyTorch, it's not that complicated as I imagined. I used to install older version of PyTorch through the direct url found in PyTorch archive. But after re-visiting the official website of PyTorch to find some information, I found that we can just simply put this argument But there are a few problems:
Besides, would you like to make GitHub Actions be able to skip builds? In my opinion, it's a nice feature when we just want to update documentation without triggering CI. Travis-CI supports this feature by adding a line |
Thanks for the help @NaleRaphael. I fixed the issues and added the "skip ci" |
In this run, there is a error "RuntimeError: Could not infer dtype of numpy.int64". But it's weird, it works fine on my machine... |
I could reproduce locally. But ya, it's weird, somehow all is fine with Pytorch 0.4.1 and Python 3.5 and 3.6; but with Python 3.7 I get the error ¯_(ツ)_/¯ |
OK, though I find a similar issue here, it seems not worthy to dig into it. |
See #42 for the discussion that led to this PR. Thanks @yongduek for bringing this issue to our attention.
@NaleRaphael would you kindly do a review of the changes here?