pytorch/xla

Pull Requests Commits

Update clip_grad_norm_ patched function signature (#2426)

malfet committed 5 years ago

Verified a158e9fe

Add warning on loss values when using XLA_USE_BF16 (#2425)

jysohn23 committed 5 years ago

Verified ef054762

updated description for code (#2417)

lezwon committed 5 years ago

Verified 9e9bc80c

changed ParallelLoader to MpDeviceLoader

lezwon committed 5 years ago

210bfe31

Skip TestArgMax/MinSameValue. (#2416)

ailzhang committed 5 years ago

Verified 9a314b71

Add section to README about persistent disk for distributed training. (#2408)

zcain117 committed 5 years ago

Verified 2c732f20

Remove hacky copy_ registration & added a backward test for einsum. (#2401)

ailzhang committed 5 years ago

Verified 057e0b70

Cleanup code to use common complete function.

davidel committed 5 years ago

4b3559ca

Move CXXABI setting into C++ tests cmake machinery so that simple builds get that as well.

davidel committed 5 years ago

0ce386bd

Disable 2 tests checking contiguity. (#2398)

ailzhang committed 5 years ago

Verified 768bbb13

Add and use API which performs explicit capture of the MultiWait shared pointer.

davidel committed 5 years ago

4d755122

Disable CXX 11 ABI when building wheels (#2390)

jysohn23 committed 5 years ago

Verified 18fa011c

Issue notify in locked section to allow use from locals.

davidel committed 5 years ago

7a2f0493

Trim TF ops. (#2391)

davidel committed 5 years ago

Verified 354b2429

Move nll_loss2d_forward/backward to optional. (#2388)

ailzhang committed 5 years ago

Verified 63ff8fb3

Handling Tensor? to c10::optional<Tensor>. (#2372)

ailzhang committed 5 years ago

Verified 5aa8dbac

Remove unused registration for einsum. (#2385)

ailzhang committed 5 years ago

Verified 00f5d9a5

Remove resize_ in the special whitelist. (#2382)

ailzhang committed 5 years ago

Verified b5b36dfc

Remove all_to_all and collective_permute from the official API. (#2384)

taylanbil committed 5 years ago

Verified 046e304a

nll_loss2d/backward lowering (#2373)

JackCaoG committed 5 years ago

Verified c4f8873d

Disable test_discontiguous_out_cumsum which checks contiguity. (#2377)

ailzhang committed 5 years ago

Verified 7fc9c3a7

Call at::_convolution instead of at::native::_convolution (#2370)

zasdfgbnm committed 5 years ago

Verified 8cefb2a9

Disable test_masked_select_discontiguous. (#2367)

ailzhang committed 5 years ago

Verified cc3d52ab

Enable test_scatter_add_non_unique_index_xla on TPU

JackCaoG committed 5 years ago

57134f5b

Add celu/celu_ to aten_tensor_ops. (#2186)

ailzhang committed 5 years ago

Verified 9b55685a

Capture device by value.

dlibenzi committed 5 years ago

fc79ec0a

Pad the input on transposed conv forward with output_padding

JackCaoG committed 5 years ago

3be91400

Avoid adding HLO to XLA graph when there is no need.

dlibenzi committed 5 years ago

c5f440b5

Make cross replica reduce token a device data instead of XLA constant.

dlibenzi committed 5 years ago

b73615c0

Use python's environment when installing xla packages (#2344)

Mistobaan committed 5 years ago

Verified a42d78c3

Newer Older