pritamdamania87
Follow
pritamdamania87
Follow
Popular repositories
-
-
pytorch Public
Forked from pytorch/pytorch
Tensors and Dynamic neural networks in Python with strong GPU acceleration
C++ 1
-
-
tensorflow Public
Forked from tensorflow/tensorflow
An Open Source Machine Learning Framework for Everyone
C++
-
-
278 contributions in the last year
Contribution activity
May 2022
Created 11 commits in 1 repository
Created 1 repository
- pritamdamania87/torchrec Python
Created a pull request in pytorch/pytorch that received 25 comments
Add sharding tests to multigpu-test.sh and fix custom operator decorator
Enabled multigpu tests. Fixed failing multigpu tests. Fixed custom operator decorator to be first preference in operator dispatch.
+87
−46
•
25
comments
Opened 11 other pull requests in 2 repositories
pytorch/pytorch
1
merged
8
closed
1
open
- [1.12] Add sharding tests to multigpu-test.sh and fix custom operator decorator
- [reland] Add sharding tests to multigpu-test.sh and fix custom operator decorator
- Validate that tensors are contiguous in ProcessGroupNCCL
- Add sharding tests to multigpu-test.sh
- Refactor operator dispatch framework across different Tensors.
- Make PartialTensot a torch.Tensor subclass
- Fix partial_tensor ops.
-
Make
biasa ReplicatedTensor in sharded linear. - Use _process_group in ReplicatedTensor and ShardedTensor.
- General fixes for ShardedTensor op framework.
pytorch/torchrec
1
closed
Reviewed 25 pull requests in 2 repositories
pytorch/pytorch
24 pull requests
- [checkpoint] Implement interop between Tensor and Sharded Tensor
- [PT-D] Enable nan_to_num op for sharded tensor
- [shard] make ShardedTensor a torch.Tensor subclass
- Add sharding tests to multigpu-test.sh and fix custom operator decorator
- Refactor operator dispatch framework across different Tensors.
- [shard] fix failed tests in sharded tensor
- [c10d] Make broadcast traceble by AoTAutograd
- [c10d] Make broadcast as a custom op
- Add sharding tests to multigpu-test.sh
- [PT-D][Sharding] Fix view op and matrix ops unit test
- [shard] make ShardedTensor a torch.Tensor subclass
- Make PartialTensot a torch.Tensor subclass
- [checkpoint] Synchronize error handling across all ranks
- [shard] add clone/detach and set requires_grad for ShardedTensor
- Rewrite ShardedTensor.gather to use dist.gather instead of gather_object
- [PT-D][Sharding] Enable more ops needed in the transformer model training
-
Make
biasa ReplicatedTensor in sharded linear. - [shard] add more tensor creation ops
- Additional ops for ShardedTensor, ReplicatedTensor and PartialTensor.
- Allow for custom sharding specs to register their own ops.
- Introduce distributed checkpoint with ShardedTensor.
- [shard] Add deepcopy for ShardedTensor
- Handle torch.memory_format serialization in TensorProperties.
- [PT-D][Sharding] Enable ops needed in the transformer model training
pytorch/examples
1 pull request
Created an issue in pytorch/pytorch that received 2 comments
TORCH_DISTRIBUTED_DEBUG should print the expected CollectiveFingerPrint
RuntimeError: Detected mismatch between colle…
2
comments