284 Pull requests merged by 6 people

• Automated Code Change

  #98603 merged Aug 10, 2025

• Automated Code Change

  #98623 merged Aug 10, 2025

• Move command buffer cudnn test from hlo level to the thunk level.

  #98597 merged Aug 9, 2025

• Properly end cache-building on Windows

  #98593 merged Aug 9, 2025

• Refactor comments in hlo_sharding.h.

  #98591 merged Aug 9, 2025

• Don't allow kBitcast for bf16 folding

  #98575 merged Aug 8, 2025

• Turn a DCHECK in a absl::FailedPreconditionError.

  #98562 merged Aug 8, 2025

• Upgrade the h5py version in requirement_lock files.

  #98583 merged Aug 8, 2025

• rename slice_index in partition_index gpu topology

  #98135 merged Aug 8, 2025

• Fix clang-format and BUILD dependency issues

  #92751 merged Aug 8, 2025

• Reduce default max_inflight_computations for TFRT GPU client.

  #98576 merged Aug 8, 2025

• Fix tsan races by **temporarily** moving sdy-round-trip-export to before other passes in MlirToXlaComputation.

  #98579 merged Aug 8, 2025

• Avoid dereferencing potentially unavailable buffer in VLOG.

  #98581 merged Aug 8, 2025

• [xla:collectives] Temporarily disable a failing test case.

  #98577 merged Aug 8, 2025

• [Windows][MinGW] Fix build with mingw

  #98533 merged Aug 8, 2025

• Add version number to logs.

  #98574 merged Aug 8, 2025

• Support extracting arg/result xla::OpShardings from sdy frontend attributes

  #97833 merged Aug 8, 2025

• [Windows] Fix include on windows cross compile

  #98570 merged Aug 8, 2025

• [NFC] Remove redundant std::unique_ptr wrapper.

  #98572 merged Aug 8, 2025

• [XLA:CPU] Add test_correctness tool.

  #98487 merged Aug 8, 2025

• [xla:cpu:onednn] Rename onednn_fusion to onednn_support

  #98517 merged Aug 8, 2025

• PR #29511: SPMD Dot Tests

  #98569 merged Aug 8, 2025

• Remove dead code related to a verifier option that was removed.

  #98559 merged Aug 8, 2025

• Automated Code Change

  #98339 merged Aug 8, 2025

• [XLA:GPU] Enable strength reduction for s32xs32->s32 dots

  #97356 merged Aug 8, 2025

• [XLA:CPU] Add more tracing to fusion compiler.

  #98491 merged Aug 8, 2025

• Check for module level tuple result sharding attributes in gspmd fallback.

  #98479 merged Aug 8, 2025

• [xla:cpu] Add a unified cc_library target for oneDNN.

  #98566 merged Aug 8, 2025

• [Windows] Fix include on windows cross compile

  #98527 merged Aug 8, 2025

• [XLA:GPU] Remove InputSlices fusion kind.

  #98558 merged Aug 8, 2025

• [Mosaic GPU] Do not call ptxas to query the latest supported PTX ISA.

  #98564 merged Aug 8, 2025

• [Windows] Fix windows build with clang

  #98530 merged Aug 8, 2025

• [XLA:CPU] Cache fusion pass manager.

  #98481 merged Aug 8, 2025

• [XLA:CPU] Add CPU version of fusion_to_mlir.

  #98565 merged Aug 8, 2025

• [XLA:GPU] Add an implementation of GetLatestPtxIsaVersion for the NvptxcompilerCompilationProvider.

  #98498 merged Aug 8, 2025

• Introduce --dump_tflite_model_dir to dump TFLite models in Delegate Test Suite (DTS)

  #97118 merged Aug 8, 2025

• Automated Code Change

  #98556 merged Aug 8, 2025

• Automated Code Change

  #98162 merged Aug 8, 2025

• PR #29270: [GPU] Bubble up mismatched buffer color from donation

  #98538 merged Aug 8, 2025

• Automated Code Change

  #98465 merged Aug 8, 2025

• [XLA:GPU][stream_executor] Introduce the GetLatestPtxIsaVersion API to CompilationProvider subclasses.

  #98493 merged Aug 8, 2025

• [XLA:CPU] Add multi-output fusion

  #98416 merged Aug 8, 2025

• * Change SchedulerCore/DefaultSchedulerCore interfaces to enable separate state creation that can be passed into ScheduleComputation

  #98348 merged Aug 8, 2025

• Move KernelNameTracer into xla/backends/gpu/profiler

  #98503 merged Aug 8, 2025

• Automated Code Change

  #98468 merged Aug 8, 2025

• Automated Code Change

  #98469 merged Aug 8, 2025

• [Windows] Fix include on windows cross compile

  #98528 merged Aug 8, 2025

• Improvements to KernelNameTracer

  #97351 merged Aug 8, 2025

• Introduce a flag to control debug info serialization for MLIR modules.

  #98514 merged Aug 8, 2025

• Automated Code Change

  #98454 merged Aug 8, 2025

• Added bfloat16 support for generating random input in TFLite utils

  #98524 merged Aug 8, 2025

• Integrate LLVM at llvm/llvm-project@6d4a0935c850

  #98523 merged Aug 8, 2025

• [Windows] Fix windows build with clang

  #98529 merged Aug 8, 2025

• [Windows] Fix include on windows cross compile

  #98526 merged Aug 8, 2025

• Add declaration for buildSortComparisonBody in StablehloOps.h to be used as a utility by other functions.

  #98353 merged Aug 7, 2025

• [XLA] Defer computation removal in HloModule, speed up DCE

  #98119 merged Aug 7, 2025

• Add unreduced sharding type.

  #98509 merged Aug 7, 2025

• [Windows] Fix include header case

  #98519 merged Aug 7, 2025

• Reverts 8bd273494f63c45db212b177b384fe2105d9139b

  #98521 merged Aug 7, 2025

• Add CreateSlicedRawBufferDest which allows copying into a slice of a raw_buffer.

  #98520 merged Aug 7, 2025

• Support integer types with fast data formatting

  #98418 merged Aug 7, 2025

• Update to disable dynamic runtime shape when compiling with TF_LITE_STATIC_MEMORY define set.

  #98440 merged Aug 7, 2025

• Introduce subclassed versions of HostOffloader for different backends.

  #97716 merged Aug 7, 2025

• Simplify coordination service identification of restarted out_of_sync workers to exclude in barrier.

  #98424 merged Aug 7, 2025

• Reduce flakiness/sensitivity of PyFunc cleanup test.

  #98516 merged Aug 7, 2025

• [Windows] Parameterize the use of msvc flags

  #98505 merged Aug 7, 2025

• [XLA:CPU] Rename codegen/math to codegen/intrinsic.

  #98512 merged Aug 7, 2025

• [NCCL] update rules_ml_toolchain archives for NCCL upgrade

  #98510 merged Aug 7, 2025

• Remove pprof_profile_builder from the XLA repository.

  #98495 merged Aug 7, 2025

• [xla:cpu:xnn] Use xnn_create_runtime_with_threadpool to create XNN runtime

  #98444 merged Aug 7, 2025

• Add unit tests on nested named computations that represent non-flat graphs.

  #98511 merged Aug 7, 2025

• Use mirrored .tar files for NVIDIA and clang archives in CI and RBE jobs.

  #98420 merged Aug 7, 2025

• Adopted f-strings for consistent Python formatting in tf_env_collect.sh

  #98400 merged Aug 7, 2025

• [XLA:GPU] update deadlocks verifier behavior to get computations from the scheduler

  #97374 merged Aug 7, 2025

• Fix FileCheck syntax.

  #98506 merged Aug 7, 2025

• Allow passing extra patches to python_init_rules

  #98431 merged Aug 7, 2025

• [XLA:CPU] Reduce Intrinsic pass boilerplate.

  #98432 merged Aug 7, 2025

• Fix a bug in converting output-operand-alias from StableHLO to HLO.

  #98497 merged Aug 7, 2025

• [XLA] Add attributes to force early/delay scheduling of nodes

  #98473 merged Aug 7, 2025

• PR #29894: Add NVIDIA Corporation to AUTHORS

  #98480 merged Aug 7, 2025

• [XLA:GPU] Update SplitK heuristic.

  #98304 merged Aug 7, 2025

• Add missing includes and update BUILD dependencies.

  #98450 merged Aug 7, 2025

• [xla:cpu:xnn] Define XnnScheduler as regular class

  #98445 merged Aug 7, 2025

• [XLA] Lazily initialize HloComputation::Builder-s in HloExtractor.

  #98492 merged Aug 7, 2025

• Reverts a9bd12a285f22d6cbff55ea48bea893d1e93cfa7

  #98457 merged Aug 7, 2025

• Support more thunks to be converted to command buffers.

  #98401 merged Aug 7, 2025

• Reverts 5e766cfc5bd2280f803e034cd14733dd03c94723

  #98490 merged Aug 7, 2025

• [tfcompile] Make some aot libs visible.

  #98308 merged Aug 7, 2025

• Introduce importOnlyUninlineableFuncCalls option to StableHLO/Sdy import pipelines.

  #98311 merged Aug 7, 2025

• [XLA] fix handling of infeed/outfeed sharding when converting mhlo to hlo

  #98020 merged Aug 7, 2025

• Fix typo in error message.

  #98430 merged Aug 7, 2025

• [XLA:GPU] Add scaled dot HLO instruction

  #97769 merged Aug 7, 2025

• Support importing all func calls and not only uninlineable ones when onlyUninlineable option is false.

  #98419 merged Aug 7, 2025

• Unify how AutotuneCacheKeys are getting constructed

  #98240 merged Aug 7, 2025

• somewhat Better logging

  #98474 merged Aug 7, 2025

• [XLA:CPU/GPU] Remove unused symbols from default indexing map.

  #98402 merged Aug 7, 2025

• Replace lite/tools/versioning:versioning with compiler/mlir/lite/tools/versioning:versioning

  #97407 merged Aug 7, 2025

• [XLA:CPU] Support benchmarking from HloSnapshot and HloUnoptimizedSnapshot.

  #98403 merged Aug 7, 2025

• [XLA:CPU][XLA:GPU] Unify optimization passes between CPU & GPU.

  #98370 merged Aug 7, 2025

• [xla] Replace deprecated symbols tsl::testing::IsOkAndHolds and tsl::testing::StatusIs.

  #98464 merged Aug 7, 2025

• Automated Code Change

  #98460 merged Aug 7, 2025

• Automated Code Change

  #98462 merged Aug 7, 2025

• Automated Code Change

  #98384 merged Aug 7, 2025

• Automated Code Change

  #98374 merged Aug 7, 2025

• Automated Code Change

  #98398 merged Aug 7, 2025

• Automated Code Change

  #98160 merged Aug 7, 2025

• Add tf2xla_dump_dir to ConfigProto.Experimental.

  #98456 merged Aug 7, 2025

• Add AllGatherPadDsSimplifier into GPU compiler pipeline,

  #97772 merged Aug 7, 2025

• Automated Code Change

  #98190 merged Aug 7, 2025

• Update version numbers for TensorFlow 2.19.1

  #98437 merged Aug 7, 2025

• Update version numbers for TensorFlow 2.20.0

  #98438 merged Aug 7, 2025

• Automated Code Change

  #98291 merged Aug 7, 2025

• [XLA:GPU] Add Pass AllGatherPadDsSimplifier to replaces a DynamicSlice(Pad(AllGather)) pattern with a more efficient sequence of HLOs, primarily using CollectivePermute and Concatenate.

  #97466 merged Aug 7, 2025

• Update RELEASE.md with 2.19.1 release notes

  #98439 merged Aug 7, 2025

• [xla:cpu:xnn] Add RAII wrappers for xnnpack types

  #98435 merged Aug 7, 2025

• Automated Code Change

  #98288 merged Aug 7, 2025

• Automated Code Change

  #98386 merged Aug 7, 2025

• [xla] Update xnnpack version

  #98449 merged Aug 7, 2025

• [XLA] Propagate constant arguments into calls.

  #98443 merged Aug 7, 2025

• [xla:cpu] NFC: Shuffle nested classes inside Thunk to improve readability

  #98448 merged Aug 7, 2025

• Upgrade rules_ml_toolchain to version with clang config

  #98423 merged Aug 7, 2025

• [xla:cpu] Detect if CpuExecutable requires XNNPACK runtime support at executable construction time

  #98446 merged Aug 7, 2025

• Fix Triton patch.

  #98441 merged Aug 7, 2025

• Add an option to accept a callback in the remat pass to allow additional operations on the rematerialized instructions.

  #98273 merged Aug 6, 2025

• [xla:cpu:xnn] Delete unused workspace

  #98428 merged Aug 6, 2025

• Integrate LLVM at llvm/llvm-project@e68a20e0b762

  #98409 merged Aug 6, 2025

• Add option to enable coordination service client to recover.

  #97460 merged Aug 6, 2025

• Update rules_ml_toolchain dependency to latest version.

  #98106 merged Aug 6, 2025

• [xla:cpu] Add Thunk::ExecuteMayBlock() flag and a ThunkExecutor optimization to parallelize processing thunks that might block execution thread

  #98426 merged Aug 6, 2025

• [XLA:CPU] Enable F64 tanh intrinsic approximation.

  #98332 merged Aug 6, 2025

• [GPU] Use uint32_t for indices in TopK CustomCall for better performance

  #98282 merged Aug 6, 2025

• [XLA] Remove accidental quadratic code in WhileLoopUnroller::IsLoopUnrollable.

  #98427 merged Aug 6, 2025

• [tsl:concurrency] Move small AsyncValue function that are on the hot path to the header file

  #98326 merged Aug 6, 2025

• [XLA:GPU] Add indexing maps for all-gather.

  #98404 merged Aug 6, 2025

• Copy pprof_profile_builder to JAX repository.

  #98415 merged Aug 6, 2025

• Checks for zero steps before iterating through a range.

  #97533 merged Aug 6, 2025

• [XLA:GPU] Sort opcode checks in indexing analysis.

  #98417 merged Aug 6, 2025

• PR #28883: [XLA:CPU][oneDNN] Add build flag to enable asynchronous support in oneDNN

  #97822 merged Aug 6, 2025

• Update rules_ml_toolchain version to enable usage of mirrored clang.tar archives.

  #98356 merged Aug 6, 2025

• Add missing illegal ops.

  #98414 merged Aug 6, 2025

• Automated Code Change

  #96839 merged Aug 6, 2025

• Import non-flat uninlineable calls into nested named computations.

  #98322 merged Aug 6, 2025

• [xla:cpu] Add support for optional XNN cost model check

  #98355 merged Aug 6, 2025

• Added use_tpu_server to the EdgeTPU settings.

  #98341 merged Aug 6, 2025

• [StableHLO/PJRT] Clean up PJRT Serialization APIs

  #98336 merged Aug 6, 2025

• [MLIR<->HLO] Split off XLA Dialect Registration

  #98105 merged Aug 6, 2025

• [SC Serving] Add XlaLocalSparseDenseMatmulOp op.

  #98255 merged Aug 6, 2025

• Update the tuple sharding scope for RecvOp to avoid copying over the entire sharding to GetTupleElement.

  #97991 merged Aug 6, 2025

• [XLA:GPU] move unstable reduction finder to OpenXLA

  #98412 merged Aug 6, 2025

• Automated Code Change

  #96838 merged Aug 6, 2025

• Automated Code Change

  #96912 merged Aug 6, 2025

• Automated Code Change

  #96862 merged Aug 6, 2025

• Automated Code Change

  #96851 merged Aug 6, 2025

• [xla:cpu] Delete unused ParallelLoopRunner

  #98327 merged Aug 6, 2025

• Sanitize constant names outside command buffer scheduling pipeline.

  #98318 merged Aug 6, 2025

• Rename ImportUnlineableFuncCalls pass into ImportFuncCalls and introduce an option to support all func calls beyond unlineable ones.

  #98245 merged Aug 6, 2025

• Automated Code Change

  #96905 merged Aug 6, 2025

• [XLA:CPU][XLA:GPU] Move shared emitter lit tests to common directory.

  #98334 merged Aug 6, 2025

• Update TFRT dependency to use revision

  #98362 merged Aug 6, 2025

• [XLA] Remove unneccessary guard in ReplaceComputations.

  #98407 merged Aug 6, 2025

• Use LazyRE2 to avoid repeated compilation of regexes in tensorflow/lite/experimental/acceleration/compatibility/canonicalize_value.cc

  #97714 merged Aug 6, 2025

• [xla:cpu:xnn] Remove pthreadpool adaptor from XLA:CPU

  #98267 merged Aug 6, 2025

• Automated Code Change

  #96856 merged Aug 6, 2025

• Integrate LLVM at llvm/llvm-project@8e7b02fc0cd5

  #98278 merged Aug 6, 2025

• [XLA:GPU] Make debugging easier when the all-reduce test fails by logging FP32 values.

  #97999 merged Aug 6, 2025

• Automated Code Change

  #98385 merged Aug 6, 2025

• Automated Code Change

  #98392 merged Aug 6, 2025

• Automated Code Change

  #98390 merged Aug 6, 2025

• Automated Code Change

  #98393 merged Aug 6, 2025

• Automated Code Change

  #98388 merged Aug 6, 2025

• Automated Code Change

  #98394 merged Aug 6, 2025

• Automated Code Change

  #98381 merged Aug 6, 2025

• Automated Code Change

  #98375 merged Aug 6, 2025

• Automated Code Change

  #98376 merged Aug 6, 2025

• Automated Code Change

  #98380 merged Aug 6, 2025

• Automated Code Change

  #98378 merged Aug 6, 2025

• [XLA:GPU] Add autotuning cache for new Autotuner.

  #98242 merged Aug 6, 2025

• [XLA:GPU] Use apply_indexing op for the contraction dimension offset.

  #98331 merged Aug 6, 2025

• Automated Code Change

  #98373 merged Aug 6, 2025

• Reverts 02058523a5a3e1773ef786f6f1b48100fb8368b4

  #98357 merged Aug 6, 2025

• Automated Code Change

  #98297 merged Aug 6, 2025

• PR #29852: Add a new HloModule clone API that can return clone context

  #98366 merged Aug 6, 2025

• Automated Code Change

  #98198 merged Aug 6, 2025

• [xla:gpu] Nested gemm fusions: allow broadcasts.

  #98236 merged Aug 6, 2025

• Add patch to fix breakage due to a recent LLVM update.

  #98364 merged Aug 6, 2025

• [Autotuner] Populate output buffers by default while profiling.

  #98316 merged Aug 6, 2025

• Disable test on ARM

  #98363 merged Aug 6, 2025

• [XLA:GPU] Remove sparsity support from HloDotInstruction.

  #97997 merged Aug 6, 2025

• Automated Code Change

  #96840 merged Aug 6, 2025

• Fix typo in XlaSparseGradientsStack shape inference

  #97857 merged Aug 6, 2025

• Check pjrt_apis is initialized in GetRegisteredPjrtApis and IsPjrtPluginInitialized.

  #97725 merged Aug 6, 2025

• Mark tensorflow/compiler/xla/python/pjrt_ifrt:pjrt_layout_serdes compatible with libtpu

  #98354 merged Aug 6, 2025

• [stablehlo] Fold SetDimensionSize ops if dimension size being set matches static result size.

  #97778 merged Aug 5, 2025

• [XLA:MSA] Speed up FixSchedule by iterating over instructions once.

  #98097 merged Aug 5, 2025

• [XLA][Numerics][HLO Value Tracking] Skip adding a recovery computation if a placeholder original value cannot be created

  #98253 merged Aug 5, 2025

• [XLA] A third attempt to bring sanity to op_name propagation.

  #98343 merged Aug 5, 2025

• Flip xla_cpu_use_xnnpack flag.

  #98325 merged Aug 5, 2025

• [XLA:SPMD] Clone the scatter reduction function.

  #98344 merged Aug 5, 2025

• [PjRt CPU] Remove legacy_memory_space_behavior from CpuClientOptions

  #98264 merged Aug 5, 2025

• Update rules_ml_toolchain to introduce CUDA 13 in the downstream projects.

  #98338 merged Aug 5, 2025

• Fix XLA linking failure on ARM

  #98337 merged Aug 5, 2025

• [XLA] Propagate constant arguments into calls.

  #98333 merged Aug 5, 2025

• [XLA:CPU] Convert tanh to Intrinsic.

  #98329 merged Aug 5, 2025

• Add unit test to show ImportUninlineableFuncCalls pass does not flatten fully.

  #98319 merged Aug 5, 2025

• [XLA:CPU] Enable new fusions by default.

  #97815 merged Aug 5, 2025

• Remove DOES_NOT_SUPPORT #ifdefs from exhaustive tests

  #98270 merged Aug 5, 2025

• PR #29765: Bump ml-dtypes from 0.5.1 to 0.5.3

  #98293 merged Aug 5, 2025

• PR #29758: Bump github/codeql-action from 3.29.4 to 3.29.5

  #98289 merged Aug 5, 2025

• PR #29204: [XLA:GPU] Remove Stream Id from command buffer

  #98286 merged Aug 5, 2025

• Remove specializations of std::is_signed.

  #98321 merged Aug 5, 2025

• Enable new autotuner for GemmAlgorithmPicker. New autotuner can be used by setting --xla_gpu_experimental_use_autotuner_pass = true.

  #98320 merged Aug 5, 2025

• Update copy thunk support in CommandBUfferConversionPass. Only device->device copies can be added to command buffer sequence.

  #98317 merged Aug 5, 2025

• [XLA:CPU] Reduce number of iterations in stateless_random_ops_test for CPU tests.

  #98312 merged Aug 5, 2025

• Add correctness and Redzone checking in Autotuner.

  #98279 merged Aug 5, 2025

• #sdy early exit if no meshes to dedup.

  #98306 merged Aug 5, 2025

• Automated Code Change

  #97896 merged Aug 5, 2025

• [XLA:GPU] Pass KernelArguments to functions instead of a span.

  #98246 merged Aug 5, 2025

• [XLA:CPU] Emulate old unrolling metadata by adding it when lowering shared loop.

  #98247 merged Aug 5, 2025

• Automated Code Change

  #98292 merged Aug 5, 2025

• PR #28919: [NVIDIA GPU] [XLA_GPU_MS_COLLECTIVE] Round-robin stream assignment for async communications

  #98290 merged Aug 5, 2025

• PR #27943: [XLA:GPU] Add support for intel gpu backend for tests

  #98281 merged Aug 5, 2025

• Automated Code Change

  #98191 merged Aug 5, 2025

• Make cudnn_sdpa_score_mod_test a xla_cc_test

  #98239 merged Aug 5, 2025

• Rename split test utils and flags to be more general.

  #98029 merged Aug 5, 2025

• [Autotuner] Extend profiler to check buffers used for profiling.

  #98093 merged Aug 5, 2025

• Make dot_decomposer handle zero-length dimensions properly

  #98173 merged Aug 5, 2025

• Automated Code Change

  #96931 merged Aug 5, 2025

• Add a couple of flag descriptions

  #98180 merged Aug 5, 2025

• Add inline_shardy_manual_computation in CallInliner.

  #98028 merged Aug 5, 2025

• [xla:pjrt] Add PjRtFuture<>::{Map,TryMap,MapTo} APIs for consistency with PjRtFuture<T>

  #98266 merged Aug 5, 2025

• Performance improvement. This was doing some blocking on the last-ref.

  #98271 merged Aug 5, 2025

• [XLA:CPU] Fix order-dependent tests with HLO dumps

  #98274 merged Aug 5, 2025

• Add traceme to socket-server for better performance traces. (Also better EAGAIN

  #98272 merged Aug 5, 2025

• Presubmits are skipping this test because it is flakey. Fix build errors.

  #98260 merged Aug 4, 2025

• [XLA:CPU] Begin conversion of F8 float conversions to xla intrinsics.

  #98156 merged Aug 4, 2025

• [tsl:concurrency] Add ScopedAsyncValue for allocating async values embedded into other objects

  #98263 merged Aug 4, 2025

• [XLA] Add some missing includes and remove unused ones.

  #97786 merged Aug 4, 2025

• Add XNN_FLAG_NO_BROADCAST to xnn_define_batch_matrix_multiply call.

  #98259 merged Aug 4, 2025

• Fix windows compilation

  #98063 merged Aug 4, 2025

• Refactor: Extract InstructionList and Item to a separate file.

  #98261 merged Aug 4, 2025

• [HLO Translate] Add sdy to registered MLIR->HLO dialects.

  #98109 merged Aug 4, 2025

• Reverts 1d2d17c095de22391cfc01690a14519c596fc183

  #98256 merged Aug 4, 2025

• XLA: fix include path

  #98062 merged Aug 4, 2025

• [xla:doc] Update sparsecore doc title

  #98254 merged Aug 4, 2025

• Add a peek method to AncestorStack.

  #97853 merged Aug 4, 2025

• [XLA] Add verification for host-transfer send/recv.

  #98244 merged Aug 4, 2025

• debug_options_parsers_test: Don't assume the same object is reused.

  #98114 merged Aug 4, 2025

• Integrate StableHLO at openxla/stablehlo@e07debd5

  #97588 merged Aug 4, 2025

• [XLA:CPU][nanort] Do not ignore ProgramShape layouts when loading proto.

  #98241 merged Aug 4, 2025

• [XLA] Inline BitcastDtypesExpander-generated calls to reduce the chance of creating bad reshape/broadcast patterns

  #98249 merged Aug 4, 2025

• Reverts 45d8a7bc5681cef4688208c48f26bb7f64e3dab1

  #98250 merged Aug 4, 2025

• [XLA:CPU] Don't use EXPECT_OK

  #98251 merged Aug 4, 2025

• Add option to enable TFRT GPU client in C API plugin.

  #97462 merged Aug 4, 2025

• Avoid undefined behavior when LRUCache keys change.

  #97834 merged Aug 4, 2025

• Integrate LLVM at llvm/llvm-project@652048ad2578

  #98033 merged Aug 4, 2025

• [XLA:CPU] Don't inline after lowering to LLVM.

  #97832 merged Aug 4, 2025

• [XLA:CPU] Add caching for new fusion kernels.

  #97761 merged Aug 4, 2025

• [XLA:CPU][XLA:GPU] Fix cached instructions.

  #98243 merged Aug 4, 2025

• Adding test for non-flat while computations and shared bodies and conditions.

  #98095 merged Aug 4, 2025

• Automated Code Change

  #96811 merged Aug 4, 2025

• Automated Code Change

  #98206 merged Aug 4, 2025

• PR #29707: [XLA:GPU] Rename cudnn sdpa tensors name

  #98238 merged Aug 4, 2025

• [XLA:GPU] Consolidate backend config to protobuf::Any.

  #97995 merged Aug 4, 2025

• Allow de-annotation of unsupported scheduling groups

  #98176 merged Aug 4, 2025

• Add _XlaShardingV2 in TPUPartitionedOps and use if present for all op's as opposed to just XlaSharding op.

  #97348 merged Aug 4, 2025

• [XLA:GPU] Migrate GemmAlgorithmPicker to new autotuner.

  #97340 merged Aug 4, 2025

• Skip call_library_for_dot for non-dot ops.

  #98177 merged Aug 4, 2025

• Clean up xnn_fusion_test.

  #98175 merged Aug 4, 2025

• [XLA:CPU] Support dumping snapshots with unoptimized HLO.

  #98082 merged Aug 4, 2025

• Automated Code Change

  #96841 merged Aug 4, 2025

• Create SymbolicExpr canonicalization tests

  #98004 merged Aug 4, 2025

• Automated Code Change

  #98224 merged Aug 4, 2025

• Add build files and fixes to get Triton tutorials to run

  #98234 merged Aug 4, 2025

• [XLA:GPU][Tiling] Port symbolic_tile_analysis to the new tile format.

  #98179 merged Aug 4, 2025

• [XLA:GPU] Remove SparseDot op from the MHLO dialect

  #97993 merged Aug 4, 2025

• Automated Code Change

  #98132 merged Aug 4, 2025

• Automated Code Change

  #98202 merged Aug 4, 2025

• [XLA] Improve hlo_live_range_test to use HLO textual form

  #97425 merged Aug 4, 2025

• Automated Code Change

  #98150 merged Aug 4, 2025

• Fix typos in documentation strings

  #98001 merged Aug 4, 2025

• Bump ubuntu from b59d215 to a08e551 in /tensorflow/tools/gcs_test

  #98065 merged Aug 4, 2025

• Bump ubuntu from 01a3ee0 to 1ec65b2 in /tensorflow/tools/tf_sig_build_dockerfiles

  #98064 merged Aug 4, 2025

• Bump the github-actions group with 2 updates

  #98066 merged Aug 4, 2025

• Automated Code Change

  #98166 merged Aug 3, 2025

153 Pull requests opened by 4 people

• Automated Code Change

  #98188 opened Aug 4, 2025

• Automated Code Change

  #98225 opened Aug 4, 2025

• [Windows] Parameterize the use of msvc flags

  #98226 opened Aug 4, 2025

• [tfcompile] Remove old runtime support in tfcompile.

  #98235 opened Aug 4, 2025

• [XLA:CPU] Remove legacy ahead of time compilation.

  #98237 opened Aug 4, 2025

• Add DeserializePjRtTopologyDescription API to PjRtCompiler and necessary plumbing at the C API and at the Plugin.

  #98248 opened Aug 4, 2025

• Support collective cancelling in TFRT GPU client.

  #98252 opened Aug 4, 2025

• Ensure L2Norm is created only when normalizing over the last axis

  #98257 opened Aug 4, 2025

• Fuse L2 normalization patterns using Mul and a keep_dims Reshape.

  #98258 opened Aug 4, 2025

• Integrate LLVM at llvm/llvm-project@39ca925ffefe

  #98262 opened Aug 4, 2025

• Reverts 5a5c6852c44a42e52d55986e5e87815c183a653f

  #98268 opened Aug 4, 2025

• Internal visibility rule change.

  #98277 opened Aug 5, 2025

• Write the ProfileOptions into XSpace

  #98315 opened Aug 5, 2025

• PR #29511: SPMD Dot Tests

  #98323 opened Aug 5, 2025

• Add sdy shardings for wrapper main while building `XlaComputation` instead of doing it at multiple separate places including `ifrt`, `tf_jit` and `tpu_mlir_compiler`.

  #98328 opened Aug 5, 2025

• Add new CachingHloEvaluator decorator.

  #98330 opened Aug 5, 2025

• Integrate LLVM at llvm/llvm-project@2e0ddbb5cd5b

  #98340 opened Aug 5, 2025

• Mark the callop as un-inlinable in `sdy_round_trip/shard_map_export`.

  #98346 opened Aug 5, 2025

• PR #26187: Simplify HloComputation::IsFusionComputation semantics.

  #98347 opened Aug 5, 2025

• [XLA:HLO Diff] Add diff codes to diff result.

  #98350 opened Aug 5, 2025

• [XLA:HLO Diff] Consolidate the diff enums.

  #98351 opened Aug 5, 2025

• Do not treat Shardy manual computation (calls and their callees) as a special case in `CallInliner`.

  #98352 opened Aug 5, 2025

• [XLA:CPU][XLA:GPU] Unify unoptimized HLO dumping snapshot flag.

  #98405 opened Aug 6, 2025

• Move `addSdyRoundTripExportPipeline` to StableHLO->HLO conversion, so the conversion will support StableHLO+Shardy->HLO, which is what JAX lowers to with shardy enabled (default).

  #98411 opened Aug 6, 2025

• Integrate LLVM at llvm/llvm-project@2e0ddbb5cd5b

  #98413 opened Aug 6, 2025

• [XLA:GPU] Add a new `xla_detect_unstable_reductions` debug option and UnstableReductionDetector pass.

  #98421 opened Aug 6, 2025

• Add GpuOnDeviceEventReceiver and GpuOnDeviceEventCollector.

  #98425 opened Aug 6, 2025

• Add validation for class_weight keys in standardize_sample_or_class_weight.

  #98429 opened Aug 6, 2025

• Update TFRT dependency to use revision

  #98433 opened Aug 6, 2025

• Reverts changelist 740124878

  #98434 opened Aug 6, 2025

• Propagate NCCL aborts into user exceptions.

  #98436 opened Aug 6, 2025

• [NCCL] Upgrade TF NCCL version to 2.27.7

  #98442 opened Aug 6, 2025

• [XLA] Add temporary workaround for building for Linux aarch64 using NVCC. Details: https://github.com/openxla/xla/issues/28256

  #98447 opened Aug 7, 2025

• Integrate LLVM at llvm/llvm-project@2e0ddbb5cd5b

  #98452 opened Aug 7, 2025

• Integrate PM Sampling into Xplane/Trace Viewer

  #98453 opened Aug 7, 2025

• Automated Code Change

  #98459 opened Aug 7, 2025

• Automated Code Change

  #98463 opened Aug 7, 2025

• Automated Code Change

  #98466 opened Aug 7, 2025

• Automated Code Change

  #98467 opened Aug 7, 2025

• Automated Code Change

  #98470 opened Aug 7, 2025

• Automated Code Change

  #98471 opened Aug 7, 2025

• Automated Code Change

  #98472 opened Aug 7, 2025

• Automated Code Change

  #98475 opened Aug 7, 2025

• Automated Code Change

  #98476 opened Aug 7, 2025

• [XLA:GPU] Make PassManager thread_local.

  #98478 opened Aug 7, 2025

• PR #26391: [NVIDIA GPU] Dynamic SPMD iteration limit for larger fast-interconnect domain

  #98482 opened Aug 7, 2025

• Automated Code Change

  #98483 opened Aug 7, 2025

• Automated Code Change

  #98484 opened Aug 7, 2025

• Automated Code Change

  #98485 opened Aug 7, 2025

• Avoid visiting nodes in ExtractVisitor that are not part of the final module.

  #98486 opened Aug 7, 2025

• Integrate LLVM at llvm/llvm-project@e8fc808bf8e7

  #98494 opened Aug 7, 2025

• PR #29157: [XLA:GPU] Add LHS schedule mode for command buffer

  #98496 opened Aug 7, 2025

• Test & demonstrate parsing of Map<string, TensorProto> fields.

  #98499 opened Aug 7, 2025

• Add feature to demonstrate TensorProto parsing

  #98500 opened Aug 7, 2025

• Add more unit tests for supporting non-flat graphs.

  #98501 opened Aug 7, 2025

• [XLA:GPU] Add autotuning cache to Autotuner, AutotunerPass and autotuner_main.

  #98502 opened Aug 7, 2025

• Flatten but not inline non-flat graphs on shardy xla pass.

  #98504 opened Aug 7, 2025

• Skip PyFuncAndEagerPyFuncTest.testCleanup due to b/144098211.

  #98507 opened Aug 7, 2025

• [XLA] Removing double calling reduce window rewriter and moving reduce window normalizer to the new place.

  #98508 opened Aug 7, 2025

• Deduplicate named computations corresponding to the same original function with the same in and out shardings.

  #98513 opened Aug 7, 2025

• ifrt-proxy google-internal changes

  #98515 opened Aug 7, 2025

• tf.math.abs returns inconsistent results for complex inf/nan inputs on CPU and GPU fix

  #98518 opened Aug 7, 2025

• [XLA:GPU] update deadlocks verifier behavior to get computations from the scheduler

  #98522 opened Aug 7, 2025

• Internal only change

  #98525 opened Aug 8, 2025

• Integrate LLVM at llvm/llvm-project@30ad2e24ab33

  #98531 opened Aug 8, 2025

• [XLA:partitioner] Support unreduced subgroup sharding.

  #98532 opened Aug 8, 2025

• Add an option to dump TfGraphToHloCompiler related call args.

  #98534 opened Aug 8, 2025

• Added bfloat16 support for generating random input in TFLite utils

  #98535 opened Aug 8, 2025

• Update a test to use API version 10. This is to prepare for adding a test with Shardy enable, which requires API version 10.

  #98536 opened Aug 8, 2025

• Integrate LLVM at llvm/llvm-project@e8fc808bf8e7

  #98537 opened Aug 8, 2025

• Automated Code Change

  #98539 opened Aug 8, 2025

• Reverts 5b351ecc0a200dea66d63b0da5f82ef7dca4fb93

  #98540 opened Aug 8, 2025

• Integrate LLVM at llvm/llvm-project@88283a6e4601

  #98541 opened Aug 8, 2025

• Update flatbuffers to 25.2.10

  #98542 opened Aug 8, 2025

• Automated Code Change

  #98543 opened Aug 8, 2025

• Automated Code Change

  #98544 opened Aug 8, 2025

• Automated Code Change

  #98545 opened Aug 8, 2025

• Automated Code Change

  #98546 opened Aug 8, 2025

• Automated Code Change

  #98547 opened Aug 8, 2025

• Automated Code Change

  #98548 opened Aug 8, 2025

• Automated Code Change

  #98549 opened Aug 8, 2025

• Automated Code Change

  #98550 opened Aug 8, 2025

• Automated Code Change

  #98551 opened Aug 8, 2025

• Automated Code Change

  #98552 opened Aug 8, 2025

• Automated Code Change

  #98553 opened Aug 8, 2025

• Automated Code Change

  #98554 opened Aug 8, 2025

• Automated Code Change

  #98555 opened Aug 8, 2025

• Move FusionDispatchPipeline to be immediately after priority fusion

  #98557 opened Aug 8, 2025

• Automated Code Change

  #98560 opened Aug 8, 2025

• Automated Code Change

  #98561 opened Aug 8, 2025

• [XLA:CPU] Implement ParallelFusionEmitter.

  #98567 opened Aug 8, 2025

• Add lift and dedup meshes in round trip import to support adding sdy shardings directly at hlo.

  #98568 opened Aug 8, 2025

• [XLA] Propagate name scopes into shardy manual computation calls during inlining.

  #98571 opened Aug 8, 2025

• Make `StreamExecutorGpuTopologyDescription` optional in `StreamExecutorGpuClient`.

  #98573 opened Aug 8, 2025

• [XLA] Create a SliceMover HLO pass.

  #98578 opened Aug 8, 2025

• Disable CoreML testing

  #98580 opened Aug 8, 2025

• Upgrade rules_android to v0.6.5.

  #98582 opened Aug 8, 2025

• Chain GpuOnDeviceEventTrace with gpu_tracer.

  #98584 opened Aug 8, 2025

• Update rules_ml_toolchain directories structure.

  #98586 opened Aug 8, 2025

• [XLA] Add an opcode for `acosh`.

  #98587 opened Aug 8, 2025

• Integrate LLVM at llvm/llvm-project@37bcd937766d

  #98588 opened Aug 8, 2025

• Integrate LLVM at llvm/llvm-project@4e5872403651

  #98589 opened Aug 8, 2025

• For variadic scatter (fusion), a copy cannot be removed if the copy's source

  #98590 opened Aug 8, 2025

• [XLA:GPU] Make matmul_perf_table_gen_main tool usable outside of tests

  #98592 opened Aug 8, 2025

• Automated Code Change

  #98594 opened Aug 9, 2025

• Regenerate configuration proto header

  #98595 opened Aug 9, 2025

• Automated Code Change

  #98596 opened Aug 9, 2025

• Automated Code Change

  #98598 opened Aug 9, 2025

• Automated Code Change

  #98599 opened Aug 9, 2025

• Automated Code Change

  #98600 opened Aug 9, 2025

• Automated Code Change

  #98601 opened Aug 9, 2025

• Automated Code Change

  #98602 opened Aug 9, 2025

• Automated Code Change

  #98604 opened Aug 9, 2025

• Automated Code Change

  #98605 opened Aug 9, 2025

• Automated Code Change

  #98606 opened Aug 9, 2025

• Automated Code Change

  #98607 opened Aug 9, 2025

• Automated Code Change

  #98608 opened Aug 9, 2025

• Automated Code Change

  #98609 opened Aug 9, 2025

• Automated Code Change

  #98610 opened Aug 9, 2025

• Automated Code Change

  #98611 opened Aug 9, 2025

• Automated Code Change

  #98612 opened Aug 9, 2025

• Automated Code Change

  #98613 opened Aug 9, 2025

• Automated Code Change

  #98614 opened Aug 9, 2025

• Automated Code Change

  #98615 opened Aug 9, 2025

• Automated Code Change

  #98616 opened Aug 9, 2025

• Automated Code Change

  #98617 opened Aug 9, 2025

• Update configuration_generated.h for use_tpu_server flag

  #98618 opened Aug 9, 2025

• Automated Code Change

  #98619 opened Aug 10, 2025

• Automated Code Change

  #98620 opened Aug 10, 2025

• Automated Code Change

  #98621 opened Aug 10, 2025

• Automated Code Change

  #98622 opened Aug 10, 2025

• Automated Code Change

  #98624 opened Aug 10, 2025

• Automated Code Change

  #98625 opened Aug 10, 2025

• Automated Code Change

  #98626 opened Aug 10, 2025

• Automated Code Change

  #98627 opened Aug 10, 2025

• Automated Code Change

  #98628 opened Aug 10, 2025

• Automated Code Change

  #98629 opened Aug 10, 2025

• Automated Code Change

  #98630 opened Aug 10, 2025

• Automated Code Change

  #98631 opened Aug 10, 2025

• Automated Code Change

  #98632 opened Aug 10, 2025

• Automated Code Change

  #98633 opened Aug 10, 2025

• Automated Code Change

  #98634 opened Aug 10, 2025

• Automated Code Change

  #98635 opened Aug 10, 2025

• Automated Code Change

  #98636 opened Aug 10, 2025

• Automated Code Change

  #98637 opened Aug 10, 2025

• Automated Code Change

  #98638 opened Aug 10, 2025

• Automated Code Change

  #98639 opened Aug 10, 2025

• Automated Code Change

  #98641 opened Aug 10, 2025

• Automated Code Change

  #98642 opened Aug 10, 2025

• FIX(tensorflow-lite) system path is prefered over TFLITE_HOST_TOOLS_DIR

  #98643 opened Aug 10, 2025

• Fix logic error in _none_to_default function

  #98644 opened Aug 10, 2025

• Improve grammar in regularizer error message

  #98645 opened Aug 10, 2025

• Fix grammar in regex_replace docstring

  #98646 opened Aug 10, 2025

34 Issues closed by 1 person

• tensorflow importing issue

  #98488 closed Aug 8, 2025

• Tensor Flow import Error

  #98359 closed Aug 6, 2025

• TypeError: __init__() missing 1 required positional argument: 'help_string'

  #98181 closed Aug 4, 2025

• ValueError: Cannot apply autograph to a function that doesn't expose a __code__ object. If this is a @tf.function, try passing f.python_function instead.

  #98182 closed Aug 4, 2025

• wrapped__BroadcastArgs_device_/job:localhost/replica:0/task:0/device:CPU:0}} Incompatible shapes: [2,5,2] vs. [4,6,5] [Op:BroadcastArgs]

  #98183 closed Aug 4, 2025

• __wrapped__MatMul_device_/job:localhost/replica:0/task:0/device:CPU:0}} Matrix size-incompatible: In[0]: [100,64], In[1]: [100,64] [Op:MatMul]

  #98184 closed Aug 4, 2025

• ensorflow.python.framework.errors_impl.DataLossError: {{function_node __wrapped__IteratorGetNext_output_types_1_device_/job:localhost/replica:0/task:0/device:CPU:0}} inflate() failed with error -3: incorrect header check [Op:IteratorGetNext]

  #98185 closed Aug 4, 2025

• I'm spamming with "fuzzing" issues

  #98208 closed Aug 4, 2025

• __wrapped__BatchMatrixDiagPart_device_/job:localhost/replica:0/task:0/device:CPU:0}} Op BatchMatrixDiagPart is not available in GraphDef version 1286. It has been removed in version 14. Use MatrixDiagPart. [Op:BatchMatrixDiagPart]

  #98223 closed Aug 4, 2025

• NotImplementedError: Op BatchIFFT2D is not available in GraphDef version 1286. It has been removed in version 15. Use IFFT2D.

  #98222 closed Aug 4, 2025

• All kernels registered for op StatelessRandomBinomial:

  #98221 closed Aug 4, 2025

• AssertionError

  #98220 closed Aug 4, 2025

• `Callable`, or one of the following values: ('lower_and_strip_punctuation', 'lower', 'strip_punctuation'). Received: None

  #98219 closed Aug 4, 2025

• Call arguments received by layer "sequence_features" (type SequenceFeatures): \u2022 features=tf.Tensor(shape=(None, 10), dtype=float32) \u2022 training=None

  #98218 closed Aug 4, 2025

• WaitForAllTasks::3409563941431950927","grpc_status":4} [type.googleapis.com/tensorflow.CoordinationServiceError=''] [type.googleapis.com/tensorflow.CoordinationServiceError='']

  #98217 closed Aug 4, 2025

• __wrapped__CheckNumerics_device_/job:localhost/replica:0/task:0/device:CPU:0}} Numerics check failed! : Tensor had NaN values [Op:CheckNumerics]

  #98216 closed Aug 4, 2025

• tensorflow.python.framework.errors_impl.InvalidArgumentError: Expected 'tf.Tensor(False, shape=(), dtype=bool)' to be true. Summarized data:

  #98215 closed Aug 4, 2025

• Registered kernels: <no registered kernels> [[ShutdownDistributedTPU]] [Op:__inference__tpu_shutdown_fn_2]

  #98214 closed Aug 4, 2025

• All kernels registered for op CrossReplicaSum: <no registered kernels> [Op:CrossReplicaSum]

  #98213 closed Aug 4, 2025

• __wrapped__SegmentSum_device_/job:localhost/replica:0/task:0/device:CPU:0}} segment ids are not increasing [Op:SegmentSum] name: segment_sum_op

  #98212 closed Aug 4, 2025

• tensorflow.python.framework.errors_impl.InvalidArgumentError: Graph execution error:

  #98211 closed Aug 4, 2025

• All kernels registered for op StatelessRandomBinomial:

  #98210 closed Aug 4, 2025

• __wrapped__SegmentMean_device_/job:localhost/replica:0/task:0/device:CPU:0}} segment ids are not increasing [Op:SegmentMean]

  #98209 closed Aug 4, 2025

• TypeError: The following are legacy tf.layers.Layers: <keras.legacy_tf_layers.core.Flatten object at 0x00000224B2131AF0>

  #98207 closed Aug 4, 2025

• Floating point exception (core dumped)

  #98205 closed Aug 4, 2025

• AssertionError

  #98204 closed Aug 4, 2025

• . Suggested fix A: If you wish to use DenseFeatures, use a non-sequence categorical_column_with_*. Suggested fix B: If you wish to create sequence input, use

  #98203 closed Aug 4, 2025

• Error: WARNING:tensorflow:From C:\Users\36132\AppData\Local\Temp\tmp0rmbbqb2.py:12: The name tf.estimator.tpu.TPUEstimatorSpec is deprecated. Please use tf.compat.v1.estimator.tpu.TPUEstimatorSpec instead.

  #98201 closed Aug 4, 2025

• ValueError: in user code:

  #98199 closed Aug 4, 2025

• ValueError: Unknown attribute '6c2540ea-8562-4d2c-a5fa-f8576b59253e' is encountered while parsing the device spec: '6c2540ea-8562-4d2c-a5fa-f8576b59253e'.

  #98195 closed Aug 4, 2025

• __wrapped__CheckNumerics_device_/job:localhost/replica:0/task:0/device:CPU:0}} Numerics check failed : Tensor had NaN values [Op:CheckNumerics]

  #98193 closed Aug 4, 2025

• assertion failed: [] [x and y not equal to tolerance rtol = Tensor(\"assert_near/rtol:0\", shape=(), dtype=float32), atol = Tensor(\"assert_near/atol:0\", shape=(), dtype=float32)] [x (assert_near/x:

  #98192 closed Aug 4, 2025

• __wrapped__InitializeTableFromDataset_device_/job:localhost/replica:0/task:0/device:CPU:0}} HashTable has different value for same key. Key apple has 0 and trying to add value 3 [Op:InitializeTableFromDataset]

  #98186 closed Aug 4, 2025

• tensorflow import error

  #98227 closed Aug 4, 2025

14 Issues opened by 10 people

• DLL load failed while importing _pywrap_tensorflow_internal: A dynamic link library (DLL) initialization routine failed. During handling of the above exception, another exception occurred:

  #98640 opened Aug 10, 2025

• Importing TF 2.20.0rc0 and PyArrow results in a `Abort trap: 6` on MacOS ARM

  #98563 opened Aug 8, 2025

• TensorFlow lite task libs does not support 16kb page alignment in Android

  #98489 opened Aug 7, 2025

• Custom Loss subclass API causes poor convergence vs wrapper function in TensorFlow 2.19.0

  #98477 opened Aug 7, 2025

• `tf.math.abs` returns inconsistent results for complex inf/nan inputs on CPU and GPU

  #98410 opened Aug 6, 2025

• tf.math.multiply for Complex inf/nan Gives Inconsistent Results on CPU for Tensors with >= 2 Elements

  #98408 opened Aug 6, 2025

• ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

  #98406 opened Aug 6, 2025

• tensorflow 2.19.0 build issues with bazel 6.5.0 on MacOS 13.6

  #98361 opened Aug 6, 2025

• `tf.raw_ops.MatrixDiag` crashes with `Aborted (core dumped)` with a 6D tensor as input on GPU

  #98349 opened Aug 5, 2025

• `tf.linalg.lu_solve` returns inconsistent output from CPU vs GPU

  #98345 opened Aug 5, 2025

• BatchNorm is not fused to Conv layer in TFlite conversion from customized QAT model?

  #98324 opened Aug 5, 2025

• Invalid `class_weight` silently ignored in `model.fit()`, no error raised

  #98283 opened Aug 5, 2025

• Inconsistent model output after save/load when using `@tf.function(jit_compile=True)`

  #98233 opened Aug 4, 2025

• Inconsistent output after model reload when using @tf.function on `call`

  #98230 opened Aug 4, 2025

50 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.

• [FIX] multi_process_runner_test times out after 2400 seconds on gfx11xx and gfx12xx GPUs

  #97680 commented on Aug 6, 2025 • 7 new comments

• Stable delegate python api

  #93850 commented on Aug 7, 2025 • 1 new comment

• GPU: Fix overflow inconsistency in cumsum on CPU/GPU

  #97693 commented on Aug 8, 2025 • 1 new comment

• Mismatch Between Quantized TFLite Layer Outputs and Expected Mathematical Values

  #97677 commented on Aug 5, 2025 • 0 new comments

• Fix stride overflow, and solution to issue number #97165

  #97260 commented on Aug 4, 2025 • 0 new comments

• Add JAX tests for deadlock verifier

  #97459 commented on Aug 7, 2025 • 0 new comments

• Add type-safe Go session config API, docs, and tests

  #97479 commented on Aug 4, 2025 • 0 new comments

• Fix: Remove dangling reference to ExperimentalTFLiteToTosaBytecode

  #97508 commented on Aug 4, 2025 • 0 new comments

• Fix: Clarify GPU compatibility check in TFLite Model Analyzer

  #97513 commented on Aug 4, 2025 • 0 new comments

• Pipe incarnations through C plugin.

  #97540 commented on Aug 7, 2025 • 0 new comments

• Use sdy front end attributes when comparing instruction identicalness

  #97585 commented on Aug 9, 2025 • 0 new comments

• Don't construct GPU cliques with stale incarnations.

  #97700 commented on Aug 7, 2025 • 0 new comments

• Add a profiler for subprocesses.

  #97705 commented on Aug 9, 2025 • 0 new comments

• [tosa] updated scatter_nd to support duplicate indices

  #97717 commented on Aug 4, 2025 • 0 new comments

• Fix: Prevent Duplicate Plugin Factory Registration (cuDNN/cuFFT/cuBLAS)

  #97724 commented on Aug 6, 2025 • 0 new comments

• [XLA:GPU] Migrate ConvAlgorithmPicker to new autotuner backend.

  #97806 commented on Aug 6, 2025 • 0 new comments

• Unify OSS and internal logging libraries.

  #97809 commented on Aug 10, 2025 • 0 new comments

• [XLA:GPU] Add scale shape verifier for ScaledDot to hlo_verifier.

  #97819 commented on Aug 8, 2025 • 0 new comments

• #sdy Export unreduced as HLOSharding instead of frontend attributes.

  #97844 commented on Aug 9, 2025 • 0 new comments

• Automated Code Change

  #97938 commented on Aug 8, 2025 • 0 new comments

• Replace sdy::kShardingRoundTripAttr with HloSharding::kShardingFrontendAttrName

  #97989 commented on Aug 7, 2025 • 0 new comments

• Automated Code Change

  #98087 commented on Aug 5, 2025 • 0 new comments

• PR #26391: [NVIDIA GPU] Dynamic SPMD iteration limit for larger fast-interconnect domain

  #98088 commented on Aug 7, 2025 • 0 new comments

• Allow propagation of per-tensor quantization parameters across tfl.broadcast_to.

  #98128 commented on Aug 8, 2025 • 0 new comments

• Create HloEvaluatorInterface, based on the methods used by InterpreterClient.

  #98130 commented on Aug 10, 2025 • 0 new comments

• Add split-phase compilation support to interpreter reference mixin.

  #98131 commented on Aug 10, 2025 • 0 new comments

• tf.no_op DAG cycle issue

  #97512 commented on Aug 5, 2025 • 0 new comments

• xprof compilation fails with gcc 14.2

  #94035 commented on Aug 5, 2025 • 0 new comments

• Add option for deterministic tie-breaking in tf.math.reduce_min

  #97688 commented on Aug 6, 2025 • 0 new comments

• Memory leak in custom_gradient decorator

  #97697 commented on Aug 6, 2025 • 0 new comments

• `tf.linalg.eigh` faces OOM on CUDA with 6D tensor and raises SIGABRT

  #97780 commented on Aug 6, 2025 • 0 new comments

• Bug in GPU delegate output tensor extraction when Split node is included

  #98103 commented on Aug 6, 2025 • 0 new comments

• TensorFlow disables SwiftUI Previews

  #95106 commented on Aug 7, 2025 • 0 new comments

• Issue while converting a model with GRU to .tflite

  #97941 commented on Aug 7, 2025 • 0 new comments

• Precision issues occur when using tf.segment_maxwith floating-point tensors of certain dimensions.

  #97805 commented on Aug 7, 2025 • 0 new comments

• EXC_BAD_ACCESS when using GemmImplUsingEigen in MacOS

  #98002 commented on Aug 7, 2025 • 0 new comments

• Installation on Grace Hopper H200 server is not GPU-enabled

  #98115 commented on Aug 7, 2025 • 0 new comments

• Cannot Convert 51 to a shape - Movenet pose classification tutorial

  #79435 commented on Aug 8, 2025 • 0 new comments

• Create a trainable tensorflow or LiteRT (with signatures) graph from a frozen tensorflow model

  #79377 commented on Aug 8, 2025 • 0 new comments

• TensorFlow on RTX 5090

  #89272 commented on Aug 8, 2025 • 0 new comments

• The model does not have any trainable weights.

  #97599 commented on Aug 8, 2025 • 0 new comments

• tf.image.combined_non_max_suppression crashed with unmatched scores

  #97672 commented on Aug 9, 2025 • 0 new comments

• lib new version not support 16kb pages in android

  #96602 commented on Aug 10, 2025 • 0 new comments

• Fix comparison functions and add unit tests

  #94484 commented on Aug 4, 2025 • 0 new comments

• [XLA] preserve Shardy shardings in SetupDerivedInstruction.

  #95637 commented on Aug 8, 2025 • 0 new comments

• Add validation for label probability distribution in softmax_cross_entropy_with_logits

  #96387 commented on Aug 4, 2025 • 0 new comments

• Fixes L2Pool implementation to not average pooling region squares

  #96599 commented on Aug 4, 2025 • 0 new comments

• #sdy Remove MHLO shardings from round-trip export

  #96640 commented on Aug 6, 2025 • 0 new comments

• Automated Code Change

  #96924 commented on Aug 7, 2025 • 0 new comments

• Add a scratch implemention of muon

  #97231 commented on Aug 5, 2025 • 0 new comments