I am wondering if folks are interested to learn more about deployment of fast.ai models, I know, I am ! Maybe it could be included as one lecture in the future fast.ai course?
P.S. And lectures on fast.ai audio, can’t wait for those as well!
2 Likes
Model deployment is covered in lesson 2 and fastai.audio lesson will be coming
1 Like
Yes, it was, but too briefly, the inquisitive minds want to know more 
Also numerous examples of how to deploy via render and other places. There’s a tutorial in the Docs as well. Every model follows the same steps for deployment so that’s why there’s not too much into it.
1 Like
I dunno, the Full Stack Deep Learning has the entire course on deployment. Maybe we can have at least one full lecture?
Plus people (me included) have many questions on testing model, i.e. normalization, resizing, etc., could be included in that lecture as well maybe? This is, of course, more invaluable for n00bs.
Everything for testing the model is done via predict(). It’s simplified for us. For images, the resizing and normalizing is done for you. As well as tabular. You can look in the source code or the documentation and it explains it.
1 Like
I am curious to know what others are doing around version control specifically their pickled models. Is it best practice to save the previous version of the pickled model along with the code? It would seem that you would want at least 1 historic model file available just in case something goes wrong during an update and you don’t want to wait for someone to run the code to train the previous model again. Any thoughts?
1 Like