Great tool, easy to learn and use.
January 24, 2025

Great tool, easy to learn and use.

Pierre Tassel | TrustRadius Reviewer
Score 9 out of 10
Vetted Review
Verified User

Overall Satisfaction with Pytorch

I use PyTorch to teach deep learning to my university students and professionally train and deploy models. Sometimes, it is done directly or through other libraries like Transformers from Hugging Face. PyTorch is very flexible and easy to write with a battery included. It offers a nice tradeoff between helpfulness and flexibility.

Pros

  • flexibility
  • Clean code, close to the algorithm.
  • Fast
  • Handles GPUs, multiple GPUs on a single machine, CPUs, and Mac.
  • Versatile, can work efficiently on text/audio/image/tabular datasets.

Cons

  • Not easy to handle TPUs.
  • Surprisingly, some industry-standard building blocks are not there (e.g., cosine lr scheduler with warmup).
  • Deployment requires additional things not there, for example, dynamic batching.
  • Fast prototyping.
  • It's not platform-dependent.
  • The pre-processing function is already present.
The big advantage of PyTorch is how close it is to the algorithm. Oftentimes, it is easier to read Pytorch code than a given paper directly. I particularly like the object-oriented approach in model definition; it makes things very clean and easy to teach to software engineers.
TensorFlow without Keras is not a pleasant experience; when using Keras, it is pretty nice, but it feels more opinionated than PyTorch; one is less free, which is not an issue in industrial settings with classic workflow but can be an issue in research settings. JAX is great but has a steep learning curve. However, it is very lovely to use, especially on TPU.

Do you think Pytorch delivers good value for the price?

Yes

Are you happy with Pytorch's feature set?

Yes

Did Pytorch live up to sales and marketing promises?

Yes

Did implementation of Pytorch go as expected?

Yes

Would you buy Pytorch again?

Yes

Everything deep learning related if not on TPU (in such case, JAX would be better suited). For LLM deployment, libraries such as vLLM would be better suited, too; otherwise, wrapping the PyTorch model with Ray is a good option.

Comments

More Reviews of Pytorch

  1. Home
  2. Machine Learning Tools
  3. Pytorch Reviews
  4. User Review of Pytorch