[pull] master from tensorflow:master - #8796
Merged
Merged
Conversation
Add EXCLUDE_FROM_ALL to add_subdirectory() calls in multiple CMake modules to prevent unnecessary compilation of third-party dependencies during default builds. This reduces build time and resource usage. - eigen.cmake - farmhash.cmake - fft2d.cmake - flatbuffers.cmake - gemmlowp.cmake - neon2sse.cmake - ruy.cmake Signed-off-by: Tushar Darote <tdarote@qti.qualcomm.com>
The shuffle() and zip() docstrings do not mention that zipping a shuffled dataset with an unshuffled one breaks element correspondence. Users hit this and think zip is re-triggering the shuffle. Added a section to shuffle() explaining the issue with a code example showing the two correct approaches (shuffle after zip, or same seed). Added a note to zip() pointing users to the shuffle docs. Fixes #70521
`_ShuffleDataset.__init__` converted `buffer_size` straight to a tensor with no upper-bound sanity check. The C++ kernel eagerly allocates a slot for every element up to `buffer_size` when the iterator is created (`std::vector<std::vector<Tensor>>(buffer_size_)` in shuffle_dataset_op.cc), so a pathologically large value reaches that allocation and crashes the process instead of raising a catchable error. Add a sanity check in the Python wrapper that rejects `buffer_size` values above ~1 billion elements with a `ValueError` before they reach the op, and add a regression test covering the reported repro. Fixes #113167
isinstance(buffer_size, int) misses NumPy integers and constant Tensor inputs, both of which bypass the check and can still reach the crashing C++ allocation. Use tensor_util.constant_value on the converted tensor instead, and add tests for np.int64 and constant Tensor buffer_size values.
prefetch compared buffer_size with AUTOTUNE to set the legacy_autotune op attribute, but for a Tensor buffer_size that comparison produces a Tensor and the op construction fails with TypeError even though the docs allow an int64 scalar tf.Tensor. Decide from the statically known value of the converted tensor instead; symbolic tensors use the non-legacy path.
Enforce shape.num_elements() == 0 when tensor_content is omitted in ParseTensorSubmessage to prevent allocating uninitialized heap memory.
PiperOrigin-RevId: 973984402
PiperOrigin-RevId: 973997845
…nPjRtClient instead. PiperOrigin-RevId: 974000504
Currently, this is not supported by the autotuner and we get the following error: RET_CHECK failure (xla/backends/gpu/runtime/collective_thunk.cc:378) params.collective_params && params.collective_params->device_assn Collective parameters and device assignment are required for collective thunk execution PiperOrigin-RevId: 974002561
PiperOrigin-RevId: 974002876
…_tpu migration layout PiperOrigin-RevId: 974005318
… length. Problem is that DynamicUpdateSliceInt4 copied input->bytes into the output buffer without bounding it against `output->bytes`. PiperOrigin-RevId: 974008610
…-allocation PiperOrigin-RevId: 974021618
…r-size PiperOrigin-RevId: 974021982
…e-index PiperOrigin-RevId: 974022071
PiperOrigin-RevId: 974030983
PiperOrigin-RevId: 974053683
…ze-validation PiperOrigin-RevId: 974053906
… extraction Imported from GitHub PR openxla/xla#47666 Like ROCm, Intel-XPU Triton targets do not support warp specialization and do not annotate functions with `ttg.total-num-warps` attribute. Hence, this PR (like ROCm) computes `ThreadDim` from the other launch information attributes. Copybara import of the project: -- e94243603c9674f6955a3331fcb3be202fcdaf9e by Akhil Goel <akhil.goel@intel.com>: Compute ThreadDim for xpu triton Merging this change closes #47666 PiperOrigin-RevId: 974067049
…HostToDeviceTransferManager. Previously, if `CommonAsyncHostToDeviceTransferManager::Create` failed during buffer creation (e.g., due to OOM or invalid arguments), definition event promises for any previously allocated buffers were fulfilled with a generic `absl::UnknownError`. This obscured the actual failure reason. Update `Create` to capture the failure `absl::Status` from the buffer creation loop and set each definition event promise with that actual status. PiperOrigin-RevId: 974091435
…ation like so: token = create_token() token, x = opt_barrier((token, x)) ys = [opt_barrier((token, y))[1] for y in ys] without needing to add dce-sinks after all the second opt-barriers. PiperOrigin-RevId: 974098204
scaling and zero point configurations for convolution operands. Update HloInstruction::CreateConvolve and HloConvolutionInstruction to support block scaling and custom scale operands. Update HLO parser and HloCreationUtils (MakeConvolveHlo) for block-scaled convolutions. Migrate HLO passes (conv_operand_swapper, convolution_group_converter, dot_as_convolution_util, space_to_batch_converter, spmd convolution_handler) to the new CreateConvolve interface. PiperOrigin-RevId: 974117253
… emulation. PiperOrigin-RevId: 974117693
PiperOrigin-RevId: 974129754
This change should be a no-op. PiperOrigin-RevId: 974137702
PiperOrigin-RevId: 974156581
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
17 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )