Build Your Own PyTorch Trainer!
HTML-код
- Опубликовано: 27 дек 2024
- In this video, I will show you how easy and simple it is to build your own #Trainer for training #PyTorch models. This will help you make all kinds of customizations you want. This video is also an intro to #Tez, which is a small, simple, to-the-point, no-nonsense trainer/wrapper that I have built. You can find it here: github.com/abh.... I will be using Tez in all future videos to keep the videos short and precise while still doing the same things. If you like Tez, do star the repository.
Please subscribe and like the video to help me keep motivated to make awesome videos like this one. :)
To buy my book, Approaching (Almost) Any Machine Learning problem, please visit: bit.ly/buyaaml
Follow me on:
Twitter: / abhi1thakur
LinkedIn: / abhi1thakur
Kaggle: kaggle.com/abh...
Tez is available here: github.com/abhishekkrthakur/tez . Don't forget to leave a star ⭐️ if you like it!
Interesting name!
Some more suggestions:
Torch: मशाल (or शोले)
Lightening = बिजली (remember Sholay)
1 of n combinations/permutations: Pyशोलेबिजली
You are increasing the pace. Nice work :)
Your videos, it's more the merrier.
🙏
Thank you sir, really such an amazing and valuable video. I think everyone here will agree with me, your channel and work are as powerful as unbelievable ( because just watching how you explain concepts and things, it has the power of an educational and motivational content, like it says: "yes it looks hard(AI, ML, DL, NLP) because we are not involved but once we are already in, it begins to flow as water do, just get in). Thanks 🙏, and all the best.
so good expline just print missing thank u so much
Ur far far far far infinity times better than
krish naik.......infact no comparison.....
Keep posting such videos
Lol, yes I agree 😂😂.
I appreciate your work Abhishek😁
16:14 loss = self(**data), what's the keyword to google to understand what self() does here? Thanks
self is an instance to a class, which is torch.nn.Module (base) in this case, when you see () being called, it's the method __call__ being called. so you can search by torch.nn.Module __call__
There must be a method defined inside superclass nn.Module, named __call__ which would call the forward method with provided argument. So when you do self(**data) it would first search __call__ inside the original class it is an instance of, then it would search it in all its parent classes according to the MRO(Method resolution order).
Is self(*args, **kwargs) keyword stuffing? It seems very useful for a data scientist but I don't think I've seen anyone else do it before.
I love this！
Great Video
Why didn't you run the code??
ColabCode and now tez
Did this really get you a subtweet from a particular developer?
No clue. Post the link and i can tell you :)
What the name of this IDE?
visual studio code
There's already tons of such libraries are out there, which are obviously much more mature and provides more functionality. So what's the point of creating our own one?
I'm asking this because once I too created my own one and spent months developing it with tons of functionality...
But at the end I just found it so unnecessary and time wasting, since libraries like pytorch-lightning, ignite etc are already out there to use.
I'm not creating a library. I'm just creating a simple pytorch trainers and some utils to make videos for my channel and use in kaggle competitions. i have no plans to compete with fully-fledged libraries. By this video, I am just creating awareness on how easy it is to make your own pytorch trainer. I hope that answers your question :)
@@abhishekkrthakur Yeah I totally got it :) As a small trainer utility its really cool 👍.