pytorch/xla

Pull Requests Commits

Make the pjrt gpu allocator configurable (#5759)

anw90 committed 1 year ago

Verified 56733fbc

Remove some unused code from `csrc/runtime` (#5785)

will-cromar committed 1 year ago

Verified 79557ccd

Implement mark_sharding as a custom op to support dynamo spmd activation sharding (#5712)

wonjoo-wj committed 1 year ago

Verified 367f47f5

Use default method in stablehlo txt/bytecode getter of StableHLOGraphModule (#5745)

lsy323 committed 1 year ago

Verified d2f389e7

Dsiable C++ test correctly (#5783)

JackCaoG committed 1 year ago

Verified d5d02306

Refactor type conversion functions out of `//torch_xla/csrc:tensor` (#5777)

will-cromar committed 1 year ago

Verified ee210bb7

Disable flaky cpp test (#5779)

JackCaoG committed 1 year ago

Verified 963204a3

fixing neuron import (#5775)

aws-kingrj committed 1 year ago

Verified bd29e796

Support autocheckpointing (#5753)

jonb377 committed 1 year ago

Verified 46643801

fix squeeze op lowering issue when dim is not in sorted order (#5751)

zpcore committed 1 year ago

Verified 21920318

Include Dynamo executation in the executation cause analysis (#5758)

JackCaoG committed 1 year ago

Verified b8b8ea86

Remove `_unsafe_index` implementation. (#5769)

ysiraichi committed 1 year ago

Verified 28c277f0

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 1 year ago

Verified a93e14bf

Transfer data directly to the device (#5752)

will-cromar committed 1 year ago

Verified 5ca36cf5

Support PreemptionSyncManager in XlaCoordinator (#5733)

jonb377 committed 1 year ago

Verified b20a082a

Correct the multinode training doc (#5747)

vanbasten23 committed 1 year ago

Verified 4038f8e7

Destroy the ComputationClient when the program exits (#5750)

will-cromar committed 1 year ago

Verified 83778f0b

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

ysiraichi committed 1 year ago

Verified 944136fa

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

JackCaoG committed 1 year ago

Verified b6a03d9b

Remove master IP discovery test for MP (#5748)

jonb377 committed 1 year ago

Verified 595ebcf0

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

jonb377 committed 1 year ago

Verified 1f0e972a

Update set_default_tensor_type to set_default_dtype (#5734)

JackCaoG committed 1 year ago

Verified 326e42bf

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

yeounoh committed 1 year ago

Verified 63ea76cd

Add doc for multinode GPU training. (#5704)

vanbasten23 committed 1 year ago

Verified b0dca120

Add Python 3.9 build for 2.1 release (#5744)

will-cromar committed 1 year ago

Verified 421ba3e4

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

lsy323 committed 1 year ago

Verified 33ff3c29

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

will-cromar committed 1 year ago

Verified 8072a446

Set --xla_latency_hiding_scheduler_rerun to 1 (#5736)

alanwaketan committed 1 year ago

Verified 47a33d0b

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

alanwaketan committed 1 year ago

Verified 4baef3c6

Newer Older