Become a sponsor to Thomas Viehmann
Hot update: ScriptTorch. If you want superfusion of PyTorch awesomeness with JIT magic, check out the link.
I do a lot of work on PyTorch, where I am a core developer and working on various bits from core computational facilities to JIT to across the board bug fixing. I also wrote some things using PyTorch, e.g. LOTranslate, a machine translation extension for LibreOffice. I blog about PyTorch work and tips and tricks.
There are some projects where there is general interest, but no-one is currently picking them up.
Past achievements in PyTorch include
- Sped up GPU BatchNorm to be among the fastest implementations
- Implemented PyTorch's native CTC Loss
- Implemented
torch.einsum - Implemented JIT optimizations for fast training for (generalized) LSTMs
- Initial version of type hints used for autocompletion
- Initial port of PyTorch to Android
I tried this once, and it didn't work out. Now, I'll link to here whenever someone asks whether things I'm creating is online.
Deep Learning with PyTorch, my book with Eli Stevens and Luca Antiga was available for free until recently.
1 sponsor has funded t-vi’s work.
My goal is to reach 10 sponsors to publish one PyTorch tip video every month.
Featured work
-
t-vi/pytorch-tvmisc
Totally Versatile Miscellanea for Pytorch
Jupyter Notebook 477 -
t-vi/AICamera
Demonstration of using Caffe2 inside an Android application.
C++ 68 -
t-vi/candlegp
Gaussian Processes in Pytorch
Jupyter Notebook 76 -
t-vi/warp-ctc
PyTorch bindings for Warp-CTC
Cuda 42 -
t-vi/pytorch
Tensors and Dynamic neural networks in Python with strong GPU acceleration
C++ 17 -
t-vi/maskrcnn-benchmark
Fast, modular reference implementation of Instance Segmentation and Object Detection algorithms in PyTorch.
Python 19