7 ms·

PyTorch – Tensor computation with strong GPU acceleration

42 points by tmlee 7y ago

abakus 7y ago
https://github.com/blue-season/pywarm https://github.com/blue-season/pywarm PyWarm is a lightweight, high-level neural network construction API for PyTorch. It enables defining all parts of NNs in the functional way.
- ru999gol 7y ago
  I don't see much of a reason you want to clutter your code with that, if you need a wrapper around pytorch use fastai not some obscure library nobody has ever heard about.
  - wodenokoto 7y ago
    Thank god you weren’t around when keras was released.
  - jsinai 7y ago
    Use fastai if you want your code to be unreadable, unmaintable and prone to mysterious bugs.
psv1 7y ago
Apart from the official tutorials what are the best resources out there for learning PyTorch?
- abiro 7y ago
  What exactly are you looking for? I think Pytorch is so ergonomic that there is no need for other resources. The only gotchas I found were in the data loading utilities.
- kingrolo 7y ago
  I really enjoyed this book for RL. https://www.packtpub.com/gb/big-data-and-business-intelligence/hands-reinforcement-learning-pytorch-10 https://www.packtpub.com/gb/big-data-and-business-intelligen... If you learn well from videos many rave about the free fast.ai courses which now use PyTorch I believe. Seems to start with image classification. http://fast.ai http://fast.ai
sandGorgon 7y ago
The #1 NLP repo on Github and the author of the most popular NLP course is moving from PyTorch to Tensorflow 2.0/Keras https://twitter.com/GokuMohandas/status/1174497967232782336 https://twitter.com/GokuMohandas/status/1174497967232782336 >We’ll still use @PyTorch but more for research lessons, I’ll post more on this decision soon! I did deliberate for several weeks on this though but ultimately it sped up development and decreased overhead for the practical lessons by a lot.
- amelius 7y ago
  > decreased overhead for the practical lessons by a lot. What do they mean by that?
  - dual_basis 7y ago
    Presumably that it was easier to use to illustrate the actual content as opposed to implementation details. This is definitely the case, at least in TensorFlow 1.x there were a lot of details which made converting from [the way we think about the problem] to [practical implementation] more cumbersome. Some of these existed for good reason (eg. performance) while others were simply architectural cruft which the benefit of hindsight makes unnecessary.
kartayyar 7y ago
This is really old, it came out in 2016. Why did you reshare it ?
baalimago 7y ago
The framework which put my trust back into frameworks. Very good!

42 points by tmlee 7y ago

https://github.com/blue-season/pywarm https://github.com/blue-season/pywarm PyWarm is a lightweight, high-level neural network construction API for PyTorch. It enables defining all parts of NNs in the functional way.

ru999gol 7y ago
I don't see much of a reason you want to clutter your code with that, if you need a wrapper around pytorch use fastai not some obscure library nobody has ever heard about.
- wodenokoto 7y ago
  Thank god you weren’t around when keras was released.
- jsinai 7y ago
  Use fastai if you want your code to be unreadable, unmaintable and prone to mysterious bugs.

psv1 7y ago

Apart from the official tutorials what are the best resources out there for learning PyTorch?

abiro 7y ago
What exactly are you looking for? I think Pytorch is so ergonomic that there is no need for other resources. The only gotchas I found were in the data loading utilities.
kingrolo 7y ago
I really enjoyed this book for RL. https://www.packtpub.com/gb/big-data-and-business-intelligence/hands-reinforcement-learning-pytorch-10 https://www.packtpub.com/gb/big-data-and-business-intelligen... If you learn well from videos many rave about the free fast.ai courses which now use PyTorch I believe. Seems to start with image classification. http://fast.ai http://fast.ai

sandGorgon 7y ago

The #1 NLP repo on Github and the author of the most popular NLP course is moving from PyTorch to Tensorflow 2.0/Keras https://twitter.com/GokuMohandas/status/1174497967232782336 https://twitter.com/GokuMohandas/status/1174497967232782336 >We’ll still use @PyTorch but more for research lessons, I’ll post more on this decision soon! I did deliberate for several weeks on this though but ultimately it sped up development and decreased overhead for the practical lessons by a lot.

amelius 7y ago
> decreased overhead for the practical lessons by a lot. What do they mean by that?
- dual_basis 7y ago
  Presumably that it was easier to use to illustrate the actual content as opposed to implementation details. This is definitely the case, at least in TensorFlow 1.x there were a lot of details which made converting from [the way we think about the problem] to [practical implementation] more cumbersome. Some of these existed for good reason (eg. performance) while others were simply architectural cruft which the benefit of hindsight makes unnecessary.

kartayyar 7y ago

This is really old, it came out in 2016. Why did you reshare it ?

baalimago 7y ago

The framework which put my trust back into frameworks. Very good!