torchtyping
tiny-cuda-nn
torchtyping | tiny-cuda-nn | |
---|---|---|
7 | 9 | |
1,337 | 3,418 | |
- | 2.4% | |
3.2 | 5.9 | |
11 months ago | about 1 month ago | |
Python | C++ | |
Apache License 2.0 | GNU General Public License v3.0 or later |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
torchtyping
-
[D] Have their been any attempts to create a programming language specifically for machine learning?
Not really an answer to your question, but there are Python packages that try to solve the problem of tensor shapes that you mentioned, e.g. https://github.com/patrick-kidger/torchtyping or https://github.com/deepmind/tensor_annotations
-
What's New in Python 3.11?
I disagree. I've had a serious attempt at array typing using variadic generics and I'm not impressed. Python's type system has numerous issues... and now they just apply to any "ArrayWithNDimensions" type as well as any "ArrayWith2Dimenensions" type.
Variadic protocols don't exist; many operations like stacking are inexpressible; the synatx is awful and verbose; etc. etc.
I've written more about this here as part of my TorchTyping project: [0]
[0] https://github.com/patrick-kidger/torchtyping/issues/37#issu...
-
Can anyone point out the mistakes in my input layer or dimension?
also https://github.com/patrick-kidger/torchtyping
-
[D] Anyone using named tensors or a tensor annotation lib productively?
FWIW I'm the author of torchtyping so happy to answer any questions about that. :) I think people are using it!
-
[D] Ideal deep learning library
The one thing I really *really* wish got more attention was named tensors and the tensor type system. Tensor misalignment errors are a constant source of silently-failing bugs. While 3rd party libraries have attempted to fill this gap, it really needs better native support. In particular it seems like bad form to me for programmers to have to remember the specific alignment and broadcasting rules, and then have to apply them to an often poorly documented order of tensor indices. I'd really like to see something like tsalib's warp operator made part of the main library and generalized to arbitrary function application, like a named-tensor version of fold. But preferably using notation closer to that of torchtyping.
-
[P] torchtyping -- documentation + runtime type checking of tensor shapes (and dtypes, ...)
Yes it does work with numerical literals! It support using integers to specify an absolute size, strings to specify names for dimensions that should all be consistently sized (and optionally also checks named tensors), "..." to indicate batch dimensions, and so on. See the full list here.
tiny-cuda-nn
-
[D] Have their been any attempts to create a programming language specifically for machine learning?
In the opposite direction from your question is a very interesting project, TinyNN all implemented as close to the metal as possible and very fast: https://github.com/NVlabs/tiny-cuda-nn
-
A CUDA-free instant NGP renderer written entirely in Python: Support real-time rendering and camera interaction and consume less than 1GB of VRAM
This repo only implemented the rendering part of the NGP but is more simple and has a lesser amount of code compared to the original (Instant-NGP and tiny-cuda-nn).
- Tiny CUDA Neural Networks: fast C++/CUDA neural network framework
- Making 3D holograms this weekend with the very “Instant” Neural Graphics Primitives by nvidia — made this volume from 100 photos taken with an old iPhone 7 Plus
- NVlabs/tiny-CUDA-nn: fast C++/CUDA neural network framework
-
Small Neural networks in Julia 5x faster than PyTorch
...a C++ library with a CUDA backend. But these high-performance building blocks might only be saturating the GPU fully if the data is large enough.
I haven't looked at implementing these things, but I imagine uf you have smaller networks and thus less data, the large building blocks may not be optimal. You may for example want to fuse some operations to reduce memory latency from repeated memory access.
In PyTorch world, there are approaches for small networks as well, there is https://github.com/NVlabs/tiny-cuda-nn - as far as I understand from the first link in the README, it makes clever use of the CUDA shared memory, which can hold all the weights of a tiny network (but not larger ones).
- [R] Instant Neural Graphics Primitives with a Multiresolution Hash Encoding (Training a NeRF takes 5 seconds!)
- Tiny CUDA Neural Networks
- Real-Time Neural Radiance Caching for Path Tracing
What are some alternatives?
jaxtyping - Type annotations and runtime checking for shape and dtype of JAX/NumPy/PyTorch/etc. arrays. https://docs.kidger.site/jaxtyping/
instant-ngp - Instant neural graphics primitives: lightning fast NeRF and more
equinox - Elegant easy-to-use neural networks + scientific computing in JAX. https://docs.kidger.site/equinox/
blis - BLAS-like Library Instantiation Software Framework
tsalib - Tensor Shape Annotation Library (numpy, tensorflow, pytorch, ...)
diffrax - Numerical differential equation solvers in JAX. Autodifferentiable and GPU-capable. https://docs.kidger.site/diffrax/
mypy - Optional static typing for Python
juliaup - Julia installer and version multiplexer
functorch - functorch is JAX-like composable function transforms for PyTorch.
RecursiveFactorization
tensor_annotations - Annotating tensor shapes using Python types
RecursiveFactorization.jl