Pinned
1,766 contributions in the last year
Less
More
Activity overview
Contribution activity
October 2021
Created 8 commits in 2 repositories
Created a pull request in pytorch/pytorch that received 7 comments
Use shared CUPTI by default
Per #57744 statically linked CUPTI causes exception handling to break on certain compiler configurations, likely because CUPTI comes with incompati…
+21
−1
•
7
comments
Opened 6 other pull requests in 3 repositories
pytorch/pytorch
2
open
2
closed
ezyang/ghstack
1
open
facebookresearch/CrypTen
1
closed
Reviewed 33 pull requests in 3 repositories
pytorch/pytorch
31 pull requests
- Add meta support to tensor range factories
- [PyTorch] Use intrusive_ptr to save space in KernelFunction
- [POC #2] functorch reimplemented using python mode stacks #66310
- Handle shared memory cases in MathBitFallback
- Fix Dispatching not considering List[Optional[Tensor]] for dispatch
-
Refactoring
aten::copy_to use dispatcher - Upgrade onnx submodule to 85546f8c44e627f8ff1181725d03cc49f675e44f
- Implement histogramdd on CPU
- replace static_cast by dynamic_cast, found by clang-tidy
-
[Draft] Introduce
torch.distributed.nn.utils.init_meta() - Update link to qnnpack in quantization doc.
- [Draft] Experiment with no-init (2)
- Fix common device computation for comparison ops.
- Add example of FiniteTensor with __torch_dispatch__
- Add uniform to onnx export
- [WIP] wrap XLATensors into Python XLA wrapper class
- Implement derivative for narrow_copy
- assert no duplicate yaml keys in codegen
- Make a python-level FunctionalTensor()
- Add InplaceOrView boxed kernel
- Add a Functionalization pass in core
-
Split matmul into
outand non-outversions - Refactoring named tensor handling
- [codegen] teach C++ reference types how to make copies of themselves
- Allow external CUDA streams to be set as current
- Some pull request reviews not shown.
pytorch/rfcs
1 pull request
pytorch/xla
1 pull request
Created an issue in pytorch/pytorch that received 4 comments
Per-overload torch.ops API
Right now, there's a many-to-one correspondence between entries in native_functions.yaml and functions in torch.ops; specifically, all overloads ge…
4
comments
Opened 6 other issues in 2 repositories
pytorch/pytorch
5
open
- tools/amd_build/build_amd.py should fail if any file fails to write
- Bazel target all_tests improperly reports failures on CPU-only (non-CUDA) build
- Give a better error message when REGISTER_DISPATCH is used in improper context
- Better (optional) tensor list handling in native kernels
- Error on unsupported key in dispatch in native_functions.yaml