pytorch/xla

Pull Requests Commits

Remove reshape lowering (#6801)

wonjoo-wj committed 1 year ago

Verified a84509b2

[Reland]Change ATEN generator argument type to const std::optional<Generator>& (#6804)

cyyever committed 1 year ago

Verified 84e7feb3

Delete torch_patches/.torch_pin (#6812)

JackCaoG committed 1 year ago

Verified 42ecc295

Make compile count in Dynamo test more general (#6802)

JackCaoG committed 1 year ago

Verified 2e5c3b64

Inline all scalars by default in export path (#6803)

lsy323 committed 1 year ago

Verified f43ac41e

Add missing import for XLA init_method (#6809)

jonb377 committed 1 year ago

Verified 34d17c2d

Update XLA pin (#6797)

yeounoh committed 1 year ago

Verified b0ceb2ba

Add cuda 12.1 nightly wheels for cp310 and cp311. (#6793)

vanbasten23 committed 1 year ago

Verified 7d8972c8

ZeRO1: Add bucketting logic to control the size of tensors for all-gather/reduce-scatter (#6025)

jeffhataws committed 1 year ago

Verified e75677f1

Update the instruction of installing the NVIDIA container toolkit (#6791)

vanbasten23 committed 1 year ago

Verified 782f05d2

[Pallas] Allow setting FlashAttention's causal mask (#6792)

alanwaketan committed 1 year ago

Verified de42834d

Add patch for unbounded dynamism (#6783)

ghpvnist committed 1 year ago

Verified 8d579f91

Use stablehlo.composite instead of stablehlo.custom_call (#6789)

mlevesquedion committed 1 year ago

Verified 3eeb15d9

Use AddBufferDonor to replace SetUpAlias for input output aliasing (#6769)

JackCaoG committed 1 year ago

Verified d70b6c4e

Add dynamism support to `aten.embedding` and `aten.split_with_sizes` (#6781)

lsy323 committed 1 year ago

Verified 6b76c7f6

[Pallas] Make a FlashAttention Wrapper (#6785)

alanwaketan committed 1 year ago

Verified fcf24b6c

Fix preserve_rng_state for activation checkpointing (#4690)

YangFei1990 committed 1 year ago

Verified 66acfebf

Remove extraneous cd (#6787)

mlevesquedion committed 1 year ago

Verified 2d0f91c7

Add dynamism support to conv1d, view, softmax (#6777)

lsy323 committed 1 year ago

Verified e9581bc9

[SPMD] auto-construct auto-sharding mesh ids (#6770)

yeounoh committed 1 year ago

Verified 968ac834

Address deprecation in torch.distributed.checkpoint (#6773)

jonb377 committed 1 year ago

Verified 73629de2

Move multislice sharding example into torch_xla (#6776)

jonb377 committed 1 year ago

Verified f1c61d32

Add fx passes to support exporting unbounded dynamism (#6653)

lsy323 committed 1 year ago

Verified 7e0d3a5a

[Pallas] Add a bfloat16 flash attention test case (#6748)

alanwaketan committed 1 year ago

Verified 1ccd6a64

[Pallas] Introduce jax_import_guard (#6747)

alanwaketan committed 1 year ago

Verified 7cf9f10a

Revert "Reland "Removing nondeterminism from partitioner eliminates one graph" (#6703)" (#6774)

JackCaoG committed 1 year ago

Verified 30a261ab

Skip `cdouble` dtype in pow test on TPU (#6772)

lsy323 committed 1 year ago

Verified bd4420c5

Fallback to CPU for dot implementation for long tensors (#6708)

bhavya01 committed 1 year ago

Verified 7145a719

Re-land: Make `as_strided_copy` materialize a new tensor with `index`. (#6697)

ysiraichi committed 1 year ago

Verified 27a7dd35

Add unit tests for a few `foo.Tensor` aten ops (#6749)

ghpvnist committed 1 year ago

Verified 6ac32233

Newer Older