All pull requests

Pull requests

• Accept RaggedTensor entries in the data of tf.debugging.Assert.

  #127966·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Pass kOverrideDefault in SigtermNotifier to prevent premature SIGTERM termination.

  #127965·copybara-service[bot] opened Sep 23, 2026Bot·

  ·
  

• Reverts changelist 985612111

  #127964·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Fix 1-ULP precision loss and optimize softmax over size-1 axis

  size:M
  #127963·RohithPariki opened Sep 23, 2026·

  ·
  1

  2

• Integrate LLVM at llvm/llvm-project@e56c2cefc3e7

  #127962·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• [XLA] Disambiguate kFusion instructions in HloCSE CseKey hash based on the fused computation

  #127961·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Prevent integer overflow in GpuGridRange iterator and RangeInitKernel

  size:S
  #127960·RohithPariki opened Sep 23, 2026·

  ·
  1

  2

• Support LRU eviction and clear_cache in FunctionCache and PolymorphicFunction

  awaiting reviewsize:M
  #127959·RohithPariki opened Sep 23, 2026·

  ·
  1

  3

• Update TODO comments on closed duplicate bugs to active canonical bugs

  #127958·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Validate boxes tensor is finite in CropAndResize gradients

  awaiting reviewsize:M
  #127957·RohithPariki opened Sep 23, 2026·

  ·
  1

  4

• Fix unsorted_segment_mean count saturation for reduced-precision dtypes

  awaiting reviewsize:S
  #127956·RohithPariki opened Sep 23, 2026·

  ·
  1

  3

• Avoid re-logging non-variable input tensors post-evaluation in calibrator

  size:M
  #127955·RohithPariki opened Sep 23, 2026·

  ·
  1

  3

• Preserve float32 subnormals in Sign kernels

  awaiting reviewsize:M
  #127954·GodlyDonuts opened Sep 23, 2026Contributor·

  ·
  1

  18

• Move some buffer assignment stuff

  #127953·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• [XLA] Use root native_custom_call_handler_allowlist in non-GPU XLA target visibility

  #127952·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Introduce ShuffleMode::MultiRotate to kShuffle instruction.

  #127951·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Fix Pow gradient when forward or intermediate power terms overflow to inf while the exact derivative is finite (e.g., tf.keras.ops.exp2(1024.0) in float64), including higher-order and mixed derivatives.

  #127950·copybara-service[bot] opened Sep 23, 2026Bot·

  ·
  1

• [IFRT] Refactor tf2hlo_test to use real PJRT-CPU IFRT client

  #127949·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Split collective_ops AOT goldens into per-arch directories

  #127948·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• [XLA:GPU] disable RTX6000PRO test for cross compile on gb200

  #127947·copybara-service[bot] opened Sep 23, 2026Bot·

  ·
  

• Return +0 from the zero branch of the xlogy/xlog1py/xdivy packet ops

  awaiting reviewsize:M
  #127946·vishwakt opened Sep 23, 2026Contributor·

  ·
  1

• [XLA] Rebuild conditional branch roots once when users move into the branches

  #127945·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• [XLA] Sort cloned instruction lists without the partial order bookkeeping

  #127944·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• PR #46865: [NVIDIA GPU] Add runtime support for CollectiveReduce

  #127943·copybara-service[bot] opened Sep 23, 2026Bot·

  ·

• Align subtract scalar coercion across execution modes

  awaiting reviewsize:S
  #127940·GodlyDonuts opened Sep 23, 2026Contributor·

  ·
  1