pytorch/xla

Pull Requests Commits

Mention amp for xla:gpu

JackCaoG committed 2 years ago

0c5f3c40

Add GPU doc to readme

JackCaoG committed 2 years ago

368ea861

Update CI test scripts to allow continuing on test failures (#4385)

wonjoo-wj committed 2 years ago

Verified cfeaedef

Update ci baseimage tag to v0.6 (#4386)

yeounoh committed 2 years ago

Verified b490eda9

Add setup for CI on TPUs (#4018)

steventk-g committed 2 years ago

Verified c4bc972b

Revert "Add aten::_cdist_forward (#4277)" (#4381)

JackCaoG committed 2 years ago

Verified fccbc6a2

Add aten::_cdist_forward (#4277)

ymwangg committed 2 years ago

Verified 3edc88f1

Improve numerical stability of torch.sigmoid (#4311)

ymwangg committed 2 years ago

Verified 453aa651

Add gpu fsdp auto_wrap test (#4378)

JackCaoG committed 2 years ago

Verified 68491026

TF pin for 12/15/2022 (#4324)

ManfeiBai committed 2 years ago

Verified 9cf2565f

Print xla_shape_ in IR too. (#4376)

alanwaketan committed 2 years ago

Verified 86bfeb61

add `auto_wrap_policy` into XLA FSDP for automatic wrapping (#4318)

ronghanghu committed 2 years ago

Verified d7d04792

Update README link to section (#4367)

JackCaoG committed 2 years ago

Verified b96c9ffe

Enable dynamo test (#4370)

JackCaoG committed 2 years ago

Verified cad848af

Enable FSDP test on gpu ci (#4372)

JackCaoG committed 2 years ago

Verified f257a203

fix type annotations in FSDP (#4371)

ronghanghu committed 2 years ago

Verified 7819db79

Add a test case for BuildInputOutputAliases (#4329)

alanwaketan committed 2 years ago

Verified 27660d79

Update test_xla_sharding.py (#4364)

wonjoo-wj committed 2 years ago

Verified 0d89710a

Fix warnings in `xrt_init.py` (#4341)

will-cromar committed 2 years ago

Verified 7881f536

Add unsharded computation test for virtual device (#4215)

steventk-g committed 2 years ago

Verified 43116740

Update to clear_all, and add back test (removed in bad merge) (#4362)

steventk-g committed 2 years ago

Verified 18caef5d

switch to pinning layout by default in FSDP to avoid data corruption in PJRT and be consistent with xm.all_reduce (#4359)

ronghanghu committed 2 years ago

Verified b357d87d

Revert "Revert "Use C10_AS_INTARRAYREF_SLOW (#4325)" (#4346)" (#4360)

JackCaoG committed 2 years ago

Verified 547ddf79

[PJRT] Fix TPU local ordinal calculation (#4334)

will-cromar committed 2 years ago

Verified 562a80d9

Temporarily remove MKL related tests (#4345)

steventk-g committed 2 years ago

Verified 09a6e7f1

Store Terraform state in shared GCS bucket. (#4323)

will-cromar committed 2 years ago

Verified fc500b47

Revert "Use C10_AS_INTARRAYREF_SLOW (#4325)" (#4346)

JackCaoG committed 2 years ago

Verified 45542f75

Use sharding hash (#4287)

steventk-g committed 2 years ago

Verified 68e5f1be

Add sharding hash to IR nodes (#4286)

steventk-g committed 2 years ago

Verified 98b788e2

Promote int to float for tan operation (consistent with Pytorch) (#4333)

steventk-g committed 2 years ago

Verified 4f375efc

Older