PyTorch
Open-source Python deep learning framework providing tensors, automatic differentiation, and GPU acceleration.
Commenters use it daily for ML work123 and like its level of abstraction,45 and it is seen as the default framework for sharing and reproducing ML results.67 The main caveat raised is that a hard PyTorch requirement shuts out developers working in other languages, such as C++.8
Why people recommend it
Daily workhorse · 3
Commenters use it every day and credit it with making ambitious ideas feasible.
“I use pytorch daily, it has made it possible for me to play with ideas that I would have only dreamed of.”
“have built many projects with PyTorch and use it daily for all my work”
Show 1 more
“I use the normal stacks you'd expect. Python, Pytorch, Spark/ray, Jupyter/Merimo, AWS, Postgres, Mathematica”
Ecosystem default · 3
Seen as the standard ML framework, recommended for sharing research and listed in job requirements.
“use good old Pytorch”
“Python has won for me: fastapi, pytorch, langchain, streamlit, and so on and on.”
Show 1 more
“a range of ML frameworks such as TensorFlow, PyTorch, or HuggingFace Transformers”
Reference implementation · 3
Used as the baseline when validating or reimplementing models and losses.
“validated it against the PyTorch version”
“The loss is mathematically equivalent to the standard KL-divergence loss available in PyTorch.”
Show 1 more
“implementing a transformer in PyTorch just to refresh my memory”
Clean abstraction · 2
Praised for the right level of abstraction compared to TensorFlow, and for matrix notation compared to NumPy.
“the more it developed, the better it was, with (at least to my mind) the perfect level of abstraction.”
“When it comes to matrix notation, I prefer PyTorch over NumPy”
Performance · 1
A from-scratch trainer using the same cuBLAS backend ran 1.5-2.5x slower than PyTorch.
“Compared to PyTorch, it's a bit slower (probably 1.5-2.5x)”
Caveats and dissent
Python lock-in · 1
A hard PyTorch requirement can exclude projects built in native languages like C++ or Go.
“because of the hard pytorch requirements”
Memory use · 1
One commenter built a lower-VRAM alternative to PyTorch's standard KL-divergence loss for long contexts.
“The loss is mathematically equivalent to the standard KL-divergence loss available in PyTorch.”
Often compared with
Who recommends it in their own words
Threads where it came up
- Leaving Meta and PyTorch 2 mentions
- Show HN: A competition for small neural networks that play strategy games 1 mention
- Chunked KL loss, running Knowledge Distillation locally in less <6GB VRAM 1 mention
- Show HN: NegativeEV – I built a tool to help people see how bad their bets are 1 mention
- Generative AI Is an Engineering Disaster 1 mention
- Ask HN: Who is hiring? (July 2026) 1 mention
Data: pytorch.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 17
“because of the hard pytorch requirements”
“The loss is mathematically equivalent to the standard KL-divergence loss available in PyTorch.”
“it builds a PyTorch model targeting the exact pitch result”
“have built many projects with PyTorch and use it daily for all my work”
“a range of ML frameworks such as TensorFlow, PyTorch, or HuggingFace Transformers”
“Compared to PyTorch, it's a bit slower (probably 1.5-2.5x)”
“use good old Pytorch”
“built a theoretical hardware framework and PyTorch simulation”
“validated it against the PyTorch version”
“Python has won for me: fastapi, pytorch, langchain, streamlit, and so on and on.”
“implementing a transformer in PyTorch just to refresh my memory”
“When it comes to matrix notation, I prefer PyTorch over NumPy”
“a range of ML frameworks such as TensorFlow, PyTorch, or HuggingFace Transformers”
“I use the normal stacks you'd expect. Python, Pytorch, Spark/ray, Jupyter/Merimo, AWS, Postgres, Mathematica”
“how jax compares to this autotuning machinery in pytorch”
“I use pytorch daily, it has made it possible for me to play with ideas that I would have only dreamed of.”
“the more it developed, the better it was, with (at least to my mind) the perfect level of abstraction.”