ezyang
Follow
ezyang
Follow
Pinned
1,841 contributions in the last year
Less
More
Activity overview
Contribution activity
August 2021
Created 7 commits in 2 repositories
Created a pull request in pytorch/pytorch that received 16 comments
Add example of FiniteTensor with __torch_dispatch__
Stack from ghstack: -> #62819 Signed-off-by: Edward Z. Yang ezyang@fb.com
+33
−0
•
16
comments
Opened 9 other pull requests in 2 repositories
pytorch/pytorch
7
closed
1
open
- Beef up comment in AccumulateType
- Implement subclass priority for __torch_dispatch__
- [skip ci] test
- Reject kwonly arguments passed positionally in torch.ops
- __torch_dispatch__: Populate kwargs dictionary with keyword-only arguments
- [POC] Dispatcher registration from Python
- Add getPyInterpreter() API
- Add support for boxed functors
facebookresearch/CrypTen
1
open
Reviewed 35 pull requests in 5 repositories
pytorch/pytorch
31 pull requests
- Add python mode
- delegate to the proper comparison operator
- Add a Functionalization pass
- [structured] Preserve computed elements from meta func to impl
- MaybeOwned page for dev wiki
- Implement subclass priority for __torch_dispatch__
- A tiny fix in MT19937RNGEngine
- [CUDA graphs] [WIP] Prototype API and documentation
- [pytorch][PR] Remove dtype from torch.Storage and use only torch.ByteStorage
- Implement NotImplemented fallback kernel and use it to replace usage in VariableType
- support optional comparisons with different but comparable types
- enhance comparison tests for c10::optional
- Add fastpath for dot and vdot when the inputs have conj bit set to True
-
Port
normkernel to structured kernels. -
Port
log_softmax_backward_datato structured kernel -
Port
log_softmaxto structured kernel - minor fixes in c10d for Windows
- Stop exporting symbols in anonymous namespaces
- Add support for boxed functors
- __torch_dispatch__: Populate kwargs dictionary with keyword-only arguments
- Add example of FiniteTensor with __torch_dispatch__
- [POC] Dispatcher registration from Python
- Add serialization support for slots and subclass getstate/setstate
- clean torch_function handling in serialization
- Update faq.rst so OOM section mentions checkpoint
- Some pull request reviews not shown.
pytorch/pytorch.github.io
1 pull request
pytorch/pytorch-probot
1 pull request
uwsampl/pytorch
1 pull request
ezyang/ghstack
1 pull request
Created an issue in pytorch/pytorch that received 19 comments
Rewrite PyTorch CUDA backend in Triton
@chsasank opened this issue in openai/triton#150 and I thought this was pretty interesting. It's true that to actually ship this in PyTorch mainlin…
19
comments
Opened 7 other issues in 2 repositories
pytorch/pytorch
6
open
- compile_commands.json doesn't contain nvcc invocations
- Half division on low precision scalar doesn't work on CPU
- Property based testing like AFL
- Don't use try-catch to handle overload resolution handling in torch.ops
- Conjugate fallback dispatch key should be per-backend
- Make KernelFunction::makeFromUnboxedFunctor infer KernelFunctor from input argument