pytorch/xla

Pull Requests Commits

Merge branch 'master' into JackCaoG/tracing_timer

JackCaoG committed 1 year ago

Verified d47c3505

Remove `_unsafe_index` implementation. (#5769)

ysiraichi committed 1 year ago

Verified 28c277f0

Delete torch_patches/.torch_pin

JackCaoG committed 1 year ago

Verified f9a5c0a2

update doc to use PJRT_DEVICE=CUDA instead of PJRT_DEVICE=GPU (#5754)

vanbasten23 committed 1 year ago

Verified f01cdb60

Revert "Transfer data directly to the device (#5752)" (#5765)

will-cromar committed 2 years ago

Verified a93e14bf

Transfer data directly to the device (#5752)

will-cromar committed 2 years ago

Verified 5ca36cf5

Record the lazy tracing time(C++) in metrics

JackCaoG committed 2 years ago

6312700e

Support PreemptionSyncManager in XlaCoordinator (#5733)

jonb377 committed 2 years ago

Verified b20a082a

Correct the multinode training doc (#5747)

vanbasten23 committed 2 years ago

Verified 4038f8e7

Destroy the ComputationClient when the program exits (#5750)

will-cromar committed 2 years ago

Verified 83778f0b

Fix XLA tensor storage device by using `XlaDeviceToAtenDevice`. (#5743)

ysiraichi committed 2 years ago

Verified 944136fa

Add tooling to explain why a graph execution happens (#5723)

JackCaoG committed 2 years ago

Verified b6a03d9b

Remove master IP discovery test for MP (#5748)

jonb377 committed 2 years ago

Verified 595ebcf0

Support SPMD through the xla:// init_method (#5706)

jonb377 committed 2 years ago

Verified 1f0e972a

Update set_default_tensor_type to set_default_dtype (#5734)

JackCaoG committed 2 years ago

Verified 326e42bf

Add information about on-going DTensor API in spmd.md (#5735)

yeounoh committed 2 years ago

Verified 63ea76cd

Add doc for multinode GPU training. (#5704)

vanbasten23 committed 2 years ago

Verified b0dca120

Add Python 3.9 build for 2.1 release (#5744)

will-cromar committed 2 years ago

Verified 421ba3e4

Mangle root scope TF variable name during tf.saved_model export (#5738)

lsy323 committed 2 years ago

Verified 33ff3c29

Merge `--pjrt_distributed` flag with `--ddp` flag. (#5732)

will-cromar committed 2 years ago

Verified 8072a446

Set --xla_latency_hiding_scheduler_rerun to 1 (#5736)

alanwaketan committed 2 years ago

Verified 47a33d0b

Revert "Don't set $TPU_LIBRARY_PATH during import (#5698)" (#5731)

alanwaketan committed 2 years ago

Verified 4baef3c6

Promote in convolution (#5727)

qihqi committed 2 years ago

Verified 47bf2eac

Fix the missing parameter error when running mp_imagenet with torchrun (#5729)

vanbasten23 committed 2 years ago

Verified 294610a1

Make the git version appears in pip (#5728)

alanwaketan committed 2 years ago

Verified ecc0e230

Ensure dist runtime is in sync before shutting down. (#5714)

vanbasten23 committed 2 years ago

Verified fa7828f5

Patch RNG for better memory utilization in dropout layers (#5710)

yeounoh committed 2 years ago

Verified 8f121c31

add op lowering for _prelu_kernel_backward (#5724)

zpcore committed 2 years ago

Verified d28b30d2

Make sure printing XLA tensor only execute the HLO once (#5721)

JackCaoG committed 2 years ago

Verified 1ad310b7

Update OpenXLA pin to 20231022 (#5720)

alanwaketan committed 2 years ago

Verified 43cd6b50

Older