hey i just saw this thing on reddit Build a Simple Deep Learning Library .. can anyone tell me if this is worth it? I think in the part 2 of fastai course we do something similar but i am not sure. Thanks in advance
Problem with this i have seen is it only goes till CNN so i am not sure it is worth spending time on
Not really. I know studying this will be a fun journey, but you are implementing something that PyTorch already has - it’s main focus is a bit too low level. In fastai, we focus on higher level stuff such as how to choose model architectures, good initialization strategies, improving training cycles and so on (mostly covered in part 2 though).
Also, on my first glance, the babygrad library can only run on CPU. Which is not very helpful.
If you want to know how automated gradient calculation work, check Karparthy’s micrograd project on github. 
what do you recommend then? should i do part2? of fastai?
Did you finish part 1 of the course? If you think you understood at least 70% of part 1, then you should definitely move on to part 2. Meanwhile, if you want to have discussions with more people that have actually finished most of the course, join our discord community server where we have friendly conversations: FastAI community (this is a 7 day temporary invite link)
1 Like