pytorch/xla

Pull Requests Commits

Scale gradients CRS to inverse of number of replicas.

dlibenzi committed 6 years ago

07f06bf1

Add narrow_copy to ATen XLA tensor

asuhan committed 6 years ago

1cc7e828

Refresh the autograd PR patch.

dlibenzi committed 6 years ago

02170d6a

Fix typo leftout by previous PR.

dlibenzi committed 6 years ago

70dfa105

Add chain_matmul to ATen XLA tensor

asuhan committed 6 years ago

09a29a3d

Propagate replication devices into XRT compile instances.

dlibenzi committed 6 years ago

eedd61ae

Prepare for setting up Nx replication mode.

dlibenzi committed 6 years ago

a61fe633

Update XLA/autograd PR patch.

dlibenzi committed 6 years ago

92b7ddf1

Added aten::clone operation.

dlibenzi committed 6 years ago

31d31fb6

Pass vector by value instead of slice to ops

asuhan committed 6 years ago

3b1e368e

Add shift operations to ATen XLA tensor

asuhan committed 6 years ago

1b4e9668

Clear the view once we set the XLA data. Added Resnet18 parallel test as well.

dlibenzi committed 6 years ago

7ea5cb99

Leave indices empty for permutation ViewInfo

asuhan committed 6 years ago

9aee1c4a

Add simple cases of as_strided and as_strided_ to ATen XLA tensor

asuhan committed 6 years ago

fe3ad982

Add in-place bitwise operations to ATen XLA tensor

asuhan committed 6 years ago

5a6c8c8e

Do not explicitly track graph size but lazily evaluate it and possibly trim.

dlibenzi committed 6 years ago

411f15a5

Added aten::threshold_ operation.

dlibenzi committed 6 years ago

63db8794

Uniformly pass at::Scalar by value

asuhan committed 6 years ago

bc11fc0b

Fix tensor.to() API.

dlibenzi committed 6 years ago

184fe590

Add aten::flip operation

vishwakftw committed 6 years ago

d3d0c701

Added first sketch of model parallelism and fixed multi-device interaction between PT/XLA and autograd.

dlibenzi committed 6 years ago

52859ad8

Added aten::cholesky operation

vishwakftw committed 6 years ago

78e92280

Add dot to ATen XLA tensor

asuhan committed 6 years ago

87f32a37

Add _argmin and _argmax to ATen XLA tensor

asuhan committed 6 years ago

3f62f05d

Fix lack of negative dimension wrap around for all operators

asuhan committed 6 years ago

cdc88836

Remove trivial XLATensor::GetCanonicalDimension

asuhan committed 6 years ago

9c3c260f

Bump relative and absolute tolerance for TestSymEig

asuhan committed 6 years ago

cf2d9754

Fix tests which uses the default XLA device directly.

dlibenzi committed 6 years ago

304e6442

Added aten::symeig operation

vishwakftw committed 6 years ago

c9632f13

Add selu, selu_, celu, celu_ to ATen XLA type

asuhan committed 6 years ago

4b213de6

Newer Older