Deployment Platform: Render ✅

abduissa · February 1, 2019, 4:08am

Update to this—I deleted the repo and re-forked it. This time I didn’t change the name of the repo in my fork. Now it works great. Very cool service!

Wonder if forking the fastai template means I can’t rename the repo…?

jeffchen72 · February 1, 2019, 8:06am

I had updated fastai on my GCP vm instance and made sure it is on version 1.0.42. For some reason, I am still stuck with the same error. I even tried using a different server on Gradient instead of GCP. Again, I followed the Back to work instructions to update fastai to 1.0.42 and generated a new export.pkl.

I verified the download link worked correctly, but again got the out of date error.

Here’s my notebook output on Gradient with fastai 1.0.42. Am I not using learn.export() correct to output the pkl file needed for Render?

pierreguillou · February 1, 2019, 2:50pm

@anurag: yes. My computer is on Windows 10. Then, I use Anaconda Prompt as terminal.
My environment:

python : 3.6.8
fastai : 1.0.42
fastprogress : 0.1.18
torch : 1.0.0
torch cuda : 9.0 / is available
torch cudnn : 7005 / is enabled
platform : Windows-10

I do not use Jupyter on Linux.

anurag · February 1, 2019, 5:09pm

It looks like there might still be some unresolved issues in fastai 1.0.42. See this comment and open issue: https://github.com/fastai/fastai/pull/1502#issuecomment-459165078

jeffchen72 · February 1, 2019, 10:10pm

Thanks, @anurag. I will step away from this for a while and come back later then. It’s good to know that I am not going crazy. Haha!

anurag · February 2, 2019, 7:23pm

You’re absolutely right! We’ve now enabled Docker caching, so it shouldn’t build the whole thing every time. This halved the build time for the example repo.

maxim.pechyonkin · February 6, 2019, 1:58pm

Today I spent some time to deploy my leafy greens classifier.

My feedback:

I like that it will automatically redeploy the app whenever I push any change to the repository.
No hassle and very easy. I’ve never deployed a model to production, I am not familiar with Docker, nor with web development. So your service was a pleasure to work with.

I will definitely recommend it.

anurag · February 6, 2019, 7:32pm

Really great to hear!

Vikash · February 10, 2019, 7:06pm

I am also getting the same error…how did you solved it??

anurag · February 10, 2019, 9:34pm

We’re waiting for the next release of the fastai library. Until then, try using a dev install using the command here: https://github.com/fastai/fastai/pull/1502#issuecomment-461227188

anurag · February 11, 2019, 11:50pm

Heads up: fastai 1.0.43 has been released, and the CUDA issues shouldn’t be a problem if you upgrade and export your .pkl file again. I’ve also updated requirements.txt for the sample repo to include the latest version of fastai and pytorch for Linux.

edwardeasling · February 12, 2019, 7:02am

I was having the same problem as @pierreguillou, @abduissa and @Vikash , and it turns out that the direct download link to the export.pkl file was not working. I was using the direct link generator in your tutorial (https://course.fast.ai/deployment_render.html) but it was generating bad links. Once I used a different link generator, my issue disappeared

chrispmaag · February 12, 2019, 8:39am

I just got Render to work on a sample shoe dataset and am loving it! The ResNet34 model works great on my Mac, but when I try to analyze a photo using my iPhone, it seems to get stuck on ‘Analyzing’. Any tips on how to make the mobile performance return a result more quickly? For reference, in Chrome on my Mac, it takes ~5-10 seconds for the model to return a result. On my phone, it seems to be stuck on the analyzing step. On mobile, I tried both Chrome and Safari browsers, and both are having the same issue.

anurag · February 12, 2019, 4:15pm

Could you post the link to the generator? I’ll update the docs.

edwardeasling · February 12, 2019, 7:02pm

https://www.wonderplugin.com/online-tools/google-drive-direct-link-generator/

marsen · February 13, 2019, 6:58am

I am using the guide to deploy example fast.ai-v3(https://github.com/render-examples/fastai-v3).And I came to this error:

chrispmaag · February 13, 2019, 7:10am

What version of the fastai library are you using? I had some issues until I updated my version with pip install --user fastai --upgrade

marsen · February 13, 2019, 7:13am

Thank you.But i am deploying it in render,i can not change something.
I am following this guild https://course.fast.ai/deployment_render.html

chrispmaag · February 13, 2019, 7:41am

Yes, I had a similar issue. My issue was Render couldn’t find fastai v 1.042 or greater (something like that). When I upgraded the library (within GCP) and exported the pickle file again, then it worked.

marsen · February 13, 2019, 8:26am

Thank you. But i didnot use GCP in this case .I solve this problem by doing this:

edit the file “requirements.txt” in the project u just forked (fastai-v3)
modify “fastai==1.0.43” to “fastai==1.0.42”
commit it.

Then u can deploy again.it works now.