Highlights
- Arctic Code Vault Contributor
- Pro
Create your own GitHub profile
Sign up for your own profile on GitHub, the best place to host code, manage projects, and build software alongside 50 million developers.
Sign up
Pinned
2,760 contributions in the last year
Contribution activity
August 2020
Created a pull request in pytorch/pytorch that received 3 comments
Rewrite of ATen code generator
Stack from ghstack:
#42629 Rewrite of ATen code generator
How to approach reviewing this diff:
The new codegen itself lives in tools/codegen. St…
+8,460
−4,118
•
3
comments
- vmap: Fix bug with x * 0.1
- Nightly Pull
- .circleci: Remove manual docker installation
- don't run deleter if it is null. issue #43166
- Update determinism documentation
- Add support for int[n]? arguments in schema
- Do not define the macro "isnan"
- Rationalize inlining of kernels into the unboxing wrapper
- Optimize linux CI build/test matrix
- update path in CI script to access ninja
- Implement Tensor.to batching rule
- Extends test_unary_ufunc.py with numerics, contiguity, domain tests
- streamline stride propagation logic in TensorIterator
- maybe minor fix to dispatch/backend_fallback_test.cpp?
- [fx] add type annotations
- Fix type annotations for torch.sparse, enable in CI
- Move Autograd to an alias dispatch key
- Issue 43057
- Implement batching rules for some unary ops
- Rename XLAPreAutograd to AutogradXLA.
- Embed torch.nn typing annotations
- [vulkan][ci] Vulkan tests running on linux build via swiftshader (added to docker)
- Enable typechecks for torch.nn.modules
- vmap: fixed to work with functools.partial
- Don't proceed into setup.py too far if Python version is unsupported
- Some pull request reviews not shown.
Created an issue in pytorch/pytorch that received 3 comments
Reduce number of stack frames used up by dispatcher
Make a debug build of PyTorch. Load up build/bin/atest in gdb. Breakpoint on at::native::add (don't worry about the symbol not being defined; it's …
3
comments
- pytorch_android_gradle_custom_build_single is still hardcoded to 09062ef-ab58-422a-b295-36c4eed6e906
- Move caffe2_onnx_main_py3_6_clang7_ubuntu16_04 off of Caffe2 image
- Codegen should warn/error when an argument that looks like an out argument isn't kwarg-only
- Bazel build has warnings
- out kwargs are sometimes inconsistent with returned named tuple field name (and in some cases, cannot be made consistent)
- TensorOptions extensibility has rusted shut
- prelu_backward, hardshrink_backward shouldn't be a method
- torch/library.h doc rendering issue
- AutoNonVariableTypeMode should clear all autograd keys, not just Autograd