Commits

Branch selector

 master

User selector

pritamdamania87

Datepicker

All time

Commit History

Commits on Jun 17, 2022

• Ensure tensors are contiguous in functional all_gather.

  Show description for 500fb24
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  500fb24

Commits on Jun 12, 2022

• Fix shard_module to appropriately deal with sub process groups.

  Show description for a81be44
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  a81be44

Commits on Jun 10, 2022

• Use appropriate dtype for sharded linear implementation.

  Show description for b9e3d72
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  b9e3d72

Commits on Jun 4, 2022

• Fuse matmul in row-wise sharded linear to have a single matmul.

  Show description for c6ca4a4
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  c6ca4a4

Commits on Jun 1, 2022

• Remove call to .contiguous() for local_shard_t.

  Show description for 5aa2ed1
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  5aa2ed1

• Use _all_gather_base and fuse matmul for sharded linear.

  Show description for 44aa4ad
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  44aa4ad

• Fuse row-wise sharded linear matmul to increase perf.

  Show description for effd270
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  effd270

Commits on May 21, 2022

• [reland] Add sharding tests to multigpu-test.sh and fix custom operator decorator (pytorch#77987)

  Show description for 37eb315
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  37eb315

Commits on May 20, 2022

• Add sharding tests to multigpu-test.sh and fix custom operator decorator (pytorch#77825)

  Show description for 8d4c8df
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  8d4c8df

Commits on May 19, 2022

• Refactor operator dispatch framework across different Tensors.

  Show description for 327d313
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  327d313

• Validate that tensors are contiguous in ProcessGroupNCCL

  Show description for b333a75
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  b333a75

Commits on May 18, 2022

• Add sharding tests to multigpu-test.sh (pytorch#77708)

  Show description for a7cf95a
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  a7cf95a

Commits on May 17, 2022

• Make PartialTensot a torch.Tensor subclass

  Show description for 068d35a
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  068d35a

• Fix partial_tensor ops.

  Show description for c83f8ee
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  c83f8ee

Commits on May 12, 2022

• Use _process_group in ReplicatedTensor and ShardedTensor.

  Show description for bdbb7fe
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  bdbb7fe

Commits on May 11, 2022

• General fixes for ShardedTensor op framework.

  Show description for b91a149
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  b91a149

Commits on May 6, 2022

• Additional ops for ShardedTensor, ReplicatedTensor and PartialTensor.

  Show description for 9e52b50
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  9e52b50

• Allow for custom sharding specs to register their own ops.

  Show description for cc685bc
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  cc685bc

Commits on Apr 26, 2022

• Fix bug where __getstate__ of DDP looks for self._replicated_tensor_module

  Show description for b26df43
  

  pritamdamania87

  authored and

  pytorchmergebot

  committed
  b26df43