@smanzoli
This comparison is dated some good years.
@Nxck2440
PyTorch is defo the correct answer as of 2025. TensorFlow has fallen behind.
@andreikulchik8280
I feel like this comparison of the two is outdated. PyTorch has also grown a vast ecosystem, has lots of cloud support, support of many platforms, including Windows, Linux, GPUs from both Nvidia and AMD, and Apple ARM chips through MPS. It has tons of optimizations, including JIT compilation and since 2.x supports compilation of dynamic graphs to static ones. It's also the default backend for HuggingFace libraries, and even Google seems to have been moving from TF to JAX. I've been working for more than 3 years with NNs for NLP, and honestly I have never had to bother with TF in research or production, so at least from my experience, PyTorch has been standard for both in recent years, at least in the context of newer projects.
@braineaterzombie3981
If you are trying anything which is recent , then pytorch is way to go. Tensorflow is good but pytorch has that flexibility
@akremgomri9085
As far as I know, pytorch integrates GPUs as well, maybe not TPUs though. But what I like the most about pytorch apart from the dynamic computation graph, is its flexibility, for instance, dataset and dataloader made my model dev much easier to create and modify as well. The idea of dataset and dataloader reminds me of design patterns in developement in the sense that code is easy to maintain and therefore, scalable.
@sneakerboyx
Google dropping support for GPU acceleration on native Windows is just diabolical
@Looki2000
TensorFlow dropped support for GPU acceleration on Windows unless you are using WSL. That sucks
@grilledcheeze101
if you want peace of mind then go with PyTorch period
@zrizzy6958
Tensorflow GPU only works on Linux or wsl, but it's still useful for me
@maxofmax1
Tbh when I watched this I thought it was made by a famous YouTuber lol
@tiberiusrubicon9261
PyTorch is much better and flexible. You can easily install it on windows, macos and linux and use it without big problems, tensorflow on windows now requires WSL to launch, on MacOS on latest silicon chips you also can easily install torch and use tith GPU acceleration from box, on the other hand flow would require separate package for it. Torch is extremely flexible for creating architectures of models, it’s so easy to create and make rules for your own layer in torch🎉🎉🎉
@user-ob5vq3jm4x
Great comparison! Both frameworks have their strengths, and it’s awesome to see how they’re evolving to make deep learning more accessible. Thanks for breaking it down!
@JohnUrbanic-m3q
This short must be from 2019 it is so outdated. TF is now the easiest to use, via Keras, and PyTorch has long since become fine for production. The have both always been "Pythonic". Was this written by AI?
@anshumbanga1347
Which one should I learn ? Have more oppurtunities in market ? and is easier
@Swastik_1214
which to learn iff i want to join hfts or any company?
@percyjw
I tried to use a GPU with tensorflow, but never got it to work. On the other hand pytorch only needed some specific GPU packages and worked
@ChinookChgdngs
aitutorialmaker AI fixes this (AI driven Tutorials). PyTorch vs. TensorFlow comparison
@thandermax
Informative 🎉🎉🎉❤
@zhaoboxu833
In fact, a lot of 'pros' of tf in the video is now not anymore unique. Pytorch has also been developed to be quite robust in industrial environments. Needless to say that above pytorch we have huggingface libraries which basically empowered all large models, so... I kinda think pytorch has become THE standard lib in the field of ML.
@IQEGO