26 Pull requests merged by 9 people

• Add docs for triton

  #7719 merged Jul 22, 2024

• Add docs for eager mode (#7700)

  #7717 merged Jul 22, 2024

• Update contributing doc after walkthrough

  #7332 merged Jul 21, 2024

• Unify the logics to check eager mode

  #7709 merged Jul 19, 2024

• Update OpenXLA pin as of 07172024

  #7715 merged Jul 19, 2024

• Update eager.md

  #7710 merged Jul 19, 2024

• Add trapz, trapezoid, div

  #7706 merged Jul 19, 2024

• Add op support from nan_to_num to movedim

  #7712 merged Jul 19, 2024

• Add tril, tile ops support

  #7711 merged Jul 19, 2024

• Change c10::optional to std::optional

  #7702 merged Jul 19, 2024

• Update docs for 2.4 release

  #7707 merged Jul 18, 2024

• internalize _parse_xla_device

  #7675 merged Jul 18, 2024

• add depreacated decorator

  #7703 merged Jul 18, 2024

• Add docs for eager mode

  #7700 merged Jul 17, 2024

• [torch_xla2] Fix some duplicate and incorrect op registrations

  #7696 merged Jul 17, 2024

• fix type annotation for 3.8

  #7701 merged Jul 17, 2024

• Support logits_soft_cap parameter in paged_attention

  #7704 merged Jul 17, 2024

• Update README.md for 2.4 release

  #7694 merged Jul 17, 2024

• update debug_single_process to false

  #7695 merged Jul 17, 2024

• Minor update to the docs

  #7691 merged Jul 16, 2024

• Update tuple type annotation

  #7699 merged Jul 16, 2024

• Add atan2, arange, addcmul, addcdiv, _softmax_backward_data, T, H

  #7681 merged Jul 16, 2024

• Optimize execution for ops that have multiple output in eager mode

  #7680 merged Jul 16, 2024

• fix git fetch for torch pin checkout

  #7690 merged Jul 15, 2024

• Lower aten::_linalg_eigh

  #7674 merged Jul 15, 2024

• Extend timeout for tpu-info test

  #7686 merged Jul 15, 2024

6 Pull requests opened by 5 people

• Add 1-layer gradient accumulation test to check aliasing

  #7692 opened Jul 16, 2024

• Update tensor.py

  #7698 opened Jul 16, 2024

• [WIP] Use `privateuseone` dispatch key

  #7705 opened Jul 17, 2024

• [torch/xla2] Lower aten_copysign and corrcoef

  #7713 opened Jul 19, 2024

• [torch_xla2] Opinfo tests for `rsqrt`, `scalar_tensor`, `i0`, `slice*`, `scalar.window.*`

  #7716 opened Jul 19, 2024

• Fix the SPMD doc strng

  #7718 opened Jul 19, 2024

7 Issues closed by 5 people

• [torchxla2] Tensor constructor capturing

  #6854 closed Jul 19, 2024

• Op info test for `movedim .. nan_to_num`

  #7556 closed Jul 19, 2024

• Op info test for `view_copy .. zeros`

  #7574 closed Jul 18, 2024

• [API Usability] Internalize `parse_xla_device`

  #7652 closed Jul 18, 2024

• Op info test for `T .. arange`

  #7688 closed Jul 16, 2024

• PyTorch/XLA should not crash runtime during int64 dot product on TPU

  #6700 closed Jul 16, 2024

• not lowered: aten::_linalg_eigh

  #6017 closed Jul 15, 2024

4 Issues opened by 4 people

• How to test on a subset of TPUs in a TPU Pod

  #7714 opened Jul 19, 2024

• Invalid version identifier in filenames of nightly builds

  #7697 opened Jul 16, 2024

• Graph dump to optimize

  #7693 opened Jul 16, 2024

• CUDA and GPU-Flavoured Docker/Container Image Missing CUDA Support

  #7689 opened Jul 15, 2024

26 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Add support for dynamic shape in dynamo

  #7676 commented on Jul 22, 2024 • 31 new comments

• Use torch_xla.experimental.compile for all examples

  #7642 commented on Jul 19, 2024 • 6 new comments

• deprecate xrt_world_size

  #7679 commented on Jul 22, 2024 • 0 new comments

• Support of AOT compilation (refine #6992)

  #7581 commented on Jul 18, 2024 • 0 new comments

• [1/N] Use std::string_view

  #7297 commented on Jul 17, 2024 • 0 new comments

• Failing Torchbench Models: tracking issue

  #5932 commented on Jul 22, 2024 • 0 new comments

• Op info test for `index_add .. linalg.diagonal`

  #7554 commented on Jul 22, 2024 • 0 new comments

• Op info test for `linalg.vecdot .. meshgrid`

  #7555 commented on Jul 21, 2024 • 0 new comments

• Op info test for `nn.functional.hardshrink .. nn.functional.huber_loss`

  #7559 commented on Jul 21, 2024 • 0 new comments

• Op info test for `nn.functional.celu .. nn.functional.glu`

  #7558 commented on Jul 21, 2024 • 0 new comments

• Op info test for `native_dropout_backward .. nn.functional.binary_cross_entropy_with_logits`

  #7557 commented on Jul 21, 2024 • 0 new comments

• 2.4 backport PR request list

  #7242 commented on Jul 19, 2024 • 0 new comments

• Op info test for `i0`

  #7447 commented on Jul 19, 2024 • 0 new comments

• Op info test for `signal.windows.hann .. slice_scatter`

  #7568 commented on Jul 19, 2024 • 0 new comments

• Op info test for `signal.windows.exponential .. signal.windows.hamming`

  #7567 commented on Jul 19, 2024 • 0 new comments

• Equivalent of get_worker_info to split an IterableDataset

  #7667 commented on Jul 19, 2024 • 0 new comments

• [torch-xla 2.1 - 2.4] when functionalization is on, there are no aliasing for gradients when using gradient accumulation

  #7174 commented on Jul 19, 2024 • 0 new comments

• Op info test for `rsqrt .. signal.windows.cosine`

  #7566 commented on Jul 18, 2024 • 0 new comments

• Op info test for `view_copy .. zeros`

  #7575 commented on Jul 18, 2024 • 0 new comments

• Is there any way to directly execute the cached computational graph

  #7682 commented on Jul 18, 2024 • 0 new comments

• Op info test for `constant_pad_nd`

  #7367 commented on Jul 17, 2024 • 0 new comments

• Sharing tensor storage (with DLPack) results in unexpected behavior.

  #7304 commented on Jul 17, 2024 • 0 new comments

• In-place operations on an DLPack aliased XLA tensor does not propagate.

  #7198 commented on Jul 16, 2024 • 0 new comments

• How to avoid compilation in a section of code?

  #7622 commented on Jul 16, 2024 • 0 new comments

• [torchbench] `moco` fails to run with CUDA OpenXLA fallback.

  #7647 commented on Jul 16, 2024 • 0 new comments

• Op info test for `atan2 .. bfloat16`

  #7545 commented on Jul 15, 2024 • 0 new comments