257 Pull requests merged by 6 people

• Fix tfrt gpu client shutdown process

  #93058 merged May 11, 2025

• [async-pjrt] Add semaphore to limit max inflight computations

  #93105 merged May 11, 2025

• Fix a bug where ToLiteral fails with a token buffer

  #93102 merged May 10, 2025

• Remove the semaphore in ExecuteHelper

  #93100 merged May 10, 2025

• [async-pjrt] Make H2D callback nonblocking

  #93095 merged May 10, 2025

• Emit a metric for unified model id of loaded models

  #93099 merged May 10, 2025

• Automated Code Change

  #93071 merged May 10, 2025

• Automated Code Change

  #93068 merged May 10, 2025

• Fall back to the CPU tensor's shape when the device buffer has a dynamic shape

  #93089 merged May 10, 2025

• Reverts changelist 714988843

  #93038 merged May 10, 2025

• Update CustomCallApiVersion for NopReturnToken in hlo_control_flow_flattening.

  #93019 merged May 10, 2025

• XLA:CPU: If configured, use more memory types and change the

  #92982 merged May 10, 2025

• Automated Code Change

  #93079 merged May 10, 2025

• Automated Code Change

  #93001 merged May 10, 2025

• Automated Code Change

  #93072 merged May 10, 2025

• Automated Code Change

  #93003 merged May 10, 2025

• Explicitly return an error when attempting to serialize >2GiB protobufs using tsl::SerializeToStringDeterministic

  #93067 merged May 10, 2025

• Provide flexibility to derived classes for specific backends w.r.t how the initial fusion worklist is formed for a current computation.

  #92913 merged May 10, 2025

• [PJRT:GPU] Remove unnecessary synchronization in donating a buffer.

  #93065 merged May 10, 2025

• Fix a crash when printing an invalid Layout.

  #93051 merged May 10, 2025

• Enable building upb with Clang.

  #92976 merged May 9, 2025

• [XLA:benchmarks] Add a workflow file to generate GHA input matrices

  #93063 merged May 9, 2025

• Add CommonPjRtClient::CreateUninitializedBuffer and remove

  #92974 merged May 9, 2025

• Short circuit StreamExecutorGpuCompiler::Compile to xla::PjRtClient::Compile before MLIR -> XlaComputation conversion

  #93030 merged May 9, 2025

• [XLA] Simplify ArrayTypeSwitch to infer the return type from the function given.

  #92439 merged May 9, 2025

• Reverts 745718a9b6d5c65f8cf387283e4d853dbe095a87

  #93054 merged May 9, 2025

• Fix race condition in tensorflow/python/data/flat_map_utils.cc.

  #92992 merged May 9, 2025

• Add an API to compilation environments to initialize all known compilation environments

  #92825 merged May 9, 2025

• Use low latency thread pool for async PjRT

  #92991 merged May 9, 2025

• Manually migrate deprecated references to strings::StrCat and strings::StrAppend.

  #93035 merged May 9, 2025

• Update some tests.

  #92968 merged May 9, 2025

• Use the stable libtpu release for TF wheel runtime, instead of a nightly one.

  #92960 merged May 9, 2025

• Cleaned up coordination service key-value store.

  #91424 merged May 9, 2025

• Remove unused "using" statement.

  #93033 merged May 9, 2025

• Simplify AbslStringify for tstring and AlphaNumBuffer.

  #93044 merged May 9, 2025

• [XLA:GPU] Deprecate PIPELINE_PARALLELISM_OPT_LEVEL_ENABLE_CYCLE_DECOMPOSER

  #90010 merged May 9, 2025

• Use the device buffer's shape to construct a literal during D2H

  #93053 merged May 9, 2025

• Fix a type in hlo_computation.h.

  #92989 merged May 9, 2025

• Reverts 69a3d55402d26a4e03b7b8cb6b45b0a8fe553707

  #93050 merged May 9, 2025

• [XLA:GPU] (NFC) Refactor CollectivePermute processing in HloVerifier.

  #93046 merged May 9, 2025

• [XLA:GPU] Add cuBLASLt autotuner backend. The backend returns configs for cuBLASLt custom calls.

  #92966 merged May 9, 2025

• Deprecate *_size() methods to reduce the Layout API surface.

  #92915 merged May 9, 2025

• Add support for literals with custom layouts in xla::PjRtStreamExecutorBuffer::ToLiteral

  #93029 merged May 9, 2025

• Updates ConvertToSolverRequest() to deduplicate node pairs, merging their costs into a single edge.

  #92975 merged May 9, 2025

• [XLA] Fail the verifier for Call instructions that have output_to_operand_aliasing set.

  #92980 merged May 9, 2025

• Remove unused xla::CompiledMemoryStats::serialized_hlo_proto

  #92918 merged May 9, 2025

• Add boolean attribute use_shardy_partitioner to XlaCallModule op.

  #92910 merged May 9, 2025

• [GPU] Replace ConditionalThunkConfig branch_count field with method.

  #93024 merged May 9, 2025

• Add unit tests for ragged_dot to reshards pass to show it supports.

  #92963 merged May 9, 2025

• [XLA:GPU] Add RaggedAllToAll to the list of ops with supported low precision inputs.

  #93037 merged May 9, 2025

• [XLA:GPU] Add absl::StatusOr to GetSupportedConfigs for Autotuning backends. StatusOr will help in writing cleaner code for early returns.

  #93025 merged May 9, 2025

• Add basic thunk deserialization logic

  #92793 merged May 9, 2025

• Fix an error in the doc of tf.strided_slice

  #92956 merged May 9, 2025

• [XLA:GPU] Use cutlass 3.8.0

  #92884 merged May 9, 2025

• Automated Code Change

  #92953 merged May 9, 2025

• [mlir][tosa] Update variable read/write names to align with v1.0 spec

  #92911 merged May 9, 2025

• Define a type alias for tsl::RCReference<Value>

  #92946 merged May 9, 2025

• [GPU] Refactor Thunk::ToProto() to return proto instead of using output param

  #92961 merged May 9, 2025

• Automated Code Change

  #92682 merged May 9, 2025

• Initialize distributed devices

  #92942 merged May 9, 2025

• [XLA:SPMD] Fix Gather/Scatter preference initialization.

  #92977 merged May 9, 2025

• Support Pinned Memory in H2D transfer.

  #92979 merged May 9, 2025

• const-correct newly-added RegisterPluggableDevicePlugin API.

  #92990 merged May 9, 2025

• AllToAll,BatchNormGrad,BatchNormTraining,BitcastConvert,Clamp,CollectiveBroadcast,Compare,Sort : Direct StableHLO to HLO Translation

  #92591 merged May 9, 2025

• Rename tf_quant ops similar to original quant ops

  #92899 merged May 9, 2025

• Integrate LLVM at llvm/llvm-project@2d287f51eff2

  #92967 merged May 8, 2025

• AllReduce,Case,If,Map,Reduce,Return,While : Direct StableHLO to HLO Translation.

  #92590 merged May 8, 2025

• Allow multi-thread scheduling for async PJRT

  #92983 merged May 8, 2025

• Don't print the shared object correspondence map.

  #92959 merged May 8, 2025

• [TF2/C++] Add TF_SKIP_LOADING_OPTIMIZED_FUNCTION_DEPENDENCIES environment variable.

  #92965 merged May 8, 2025

• Added new GroupExecute API to XLA GPU collectives.

  #92964 merged May 8, 2025

• Use "GetNumResourcesPerInstruction" instead of "GetResourcesFromInstruction" when computing resource usage for an annotation group since the former considers resources across call hierarchies while the latter does not.

  #92971 merged May 8, 2025

• Branch quantization/satblehlo:test_passes to use TFQuantDialect

  #92138 merged May 8, 2025

• Load python rules instead of using native rules.

  #92746 merged May 8, 2025

• Replaced the GreedySubGraphExactMatcher with a new GreedyTopDownMatcher that employs stricter matching criteria. This new matcher only propagates mappings between nodes that have the same number of children, and whose corresponding children (in the same position) share the same type.

  #92820 merged May 8, 2025

• Add benchmark for Shape assignment.

  #92824 merged May 8, 2025

• Remove extraneous onednn patch

  #92970 merged May 8, 2025

• [XLA:CPU] Allow users to pass device assignment to NanoRt

  #92948 merged May 8, 2025

• Fix TFLite Compiler warnings in Chromium build

  #92908 merged May 8, 2025

• [tosa] Fix member access post-MLIR upstream 0c61b24

  #92969 merged May 8, 2025

• [XLA][runtime] Add strongly typed xla specific GlobalDeviceId and LocalDeviceId.

  #92951 merged May 8, 2025

• Fix logging format string not referring to all passed variables.

  #92919 merged May 8, 2025

• Branch quantization/stablehlo/cc:pre_calibration to use TFQuantDialect

  #92909 merged May 8, 2025

• Use LoadedExecutableRef instead of std::shared_ptr<LoadedExecutable>

  #92944 merged May 8, 2025

• Use ArrayRef instead of tsl::RCReference<Array>

  #92943 merged May 8, 2025

• Reverts 4e2dfdbb4e82695f868a90e823f9a8027ba48a4d

  #92629 merged May 8, 2025

• [Cleanup] Cleanup whitespace

  #92741 merged May 8, 2025

• [XLA:GPU] Remove CollectivePermuteValidIterationAnnotator pass

  #89923 merged May 8, 2025

• Flatten the result tuples while converting HLO to StableHLO

  #92312 merged May 8, 2025

• [XLA:GPU][Emitters] Use concatenate emitter for 4-bits datatypes

  #91916 merged May 8, 2025

• Automated Code Change

  #92950 merged May 8, 2025

• [XLA:CPU] Set fp-contract to on rather than fast for DUCC

  #92887 merged May 8, 2025

• #sdy export reduce_scatter collective op in stablehlo export.

  #92551 merged May 8, 2025

• Use transpose emitter for transposes with 4-bits datatypes

  #92777 merged May 8, 2025

• PR #25635: Add parsing debug options support

  #92940 merged May 8, 2025

• PR #26183: [XLA:GPU] Add support for SM103a and SM121a architectures (Blackwell)

  #92881 merged May 8, 2025

• Add proto serialization for GPU CopyThunk

  #92808 merged May 8, 2025

• Add To/FromProto() to BufferAllocation::Slice

  #92897 merged May 8, 2025

• [XLA:GPU] Add triton support test for stochastic convert op

  #92939 merged May 8, 2025

• Added min/max reductions on XNNPACK delegate level.

  #92782 merged May 8, 2025

• Automated Code Change

  #92931 merged May 8, 2025

• Re-enable test that is now passing.

  #92888 merged May 8, 2025

• Automated Code Change

  #92938 merged May 8, 2025

• Automated Code Change

  #92671 merged May 8, 2025

• Fixed ppc64le build issues

  #92769 merged May 8, 2025

• Remove support for sparse dimension's ordered attribute.

  #92914 merged May 8, 2025

• Integrate StableHLO at openxla/stablehlo@630c315b

  #92907 merged May 8, 2025

• Internal dir restructure

  #90497 merged May 8, 2025

• Check whether createStream was successful

  #92729 merged May 8, 2025

• Get rid of HloModuleProto from CompiledMemoryStats

  #92748 merged May 7, 2025

• Add traceme while borrowing streams

  #92916 merged May 7, 2025

• Adds checkpoint test for tpu_embedding_v3

  #92831 merged May 7, 2025

• [IFRT] Add simple string array tests to IFRT common array_impl_test

  #92823 merged May 7, 2025

• [ReplicaGroupV2] Generate all-to-all collectives using IotaReplicaGroupList format.

  #92755 merged May 7, 2025

• Add a bool flag to UpdateCompileOptions to specify whether the addressable device lookup is required for compilation.

  #92747 merged May 7, 2025

• Use ArrayRef instead of tsl::RCReference<Array>

  #92898 merged May 7, 2025

• Clarify the difference between internal and external version.

  #92830 merged May 7, 2025

• A simple constant folder for case ops

  #92812 merged May 7, 2025

• Add strings::LegacyPrecision to TSL.

  #92890 merged May 7, 2025

• Force disable build time Python pyc generation.

  #92905 merged May 7, 2025

• This CL has no functional changes, but improved code structure:

  #92091 merged May 7, 2025

• Adding an option to initialize Pluggable Device Plugin using function pointers instead of a DSO

  #92816 merged May 7, 2025

• Fix missing cuda graph launch call back event in new version of CUDA

  #92904 merged May 7, 2025

• Refactor BFS function to accept distance parameter

  #92444 merged May 7, 2025

• Branch quantization/stablehlo/cc:post_calibration to use TFQuantDialect

  #92889 merged May 7, 2025

• Introduce new CompilePhaseHloRunnerPjRt and ExecutePhaseHloRunnerPjRt.

  #92440 merged May 7, 2025

• Branch quantization/tensorflow:quantize_preprocess to use TFQuantDialect

  #92886 merged May 7, 2025

• LogScheduleStatistics is very slow. Lower it's level from 1 to 3.

  #89198 merged May 7, 2025

• [XLA:benchmarks] Add --workflow_type to the binary to generate GHA input matrices by workflow types

  #92903 merged May 7, 2025

• Fix HloRunnerPjRt DeserializeExecutable.

  #92902 merged May 7, 2025

• Migration fuse_convolution_pass to tf_attrs_and_constraints

  #92558 merged May 7, 2025

• Refactor the DiffSummary log print.

  #92628 merged May 7, 2025

• Branch quantization/stablehlo/cc:tf_pass_pipeline to use TFQuantDialect

  #92432 merged May 7, 2025

• Add an API to the URL generator for generating URLs without selecting nodes.

  #92725 merged May 7, 2025

• [IFRT] Treat Maximal HloSharding on a single device as fully replicated

  #92810 merged May 7, 2025

• Clean up LoadedExecutable::Delete and LoadedExecutable::IsDeleted

  #92791 merged May 7, 2025

• Remove support for sparse dimension's unique attribute.

  #92753 merged May 7, 2025

• Support memory space in underlying buffer allocation.

  #92821 merged May 7, 2025

• Add an optimization pattern to convert a fully_connected op with a weight with last dimension of 1 to a broadcasting mul op.

  #92893 merged May 7, 2025

• Support types that have AbslStringify, but not operator<< in the additional output of TF_RET_CHECK.

  #92720 merged May 7, 2025

• [JAX] Add a test for multiprocess shard_map in McJAX with non-participating hosts. Update handling of device memories in PyDeviceList to support this.

  #92737 merged May 7, 2025

• [XLA:benchmarks] Add a binary to generate GHA input matrices to trigger benchmark runs

  #92807 merged May 7, 2025

• Remove Shape::ShortDebugString() to reduce the Shape API surface.

  #92565 merged May 7, 2025

• Only delegate 4 bit weights which are per channel quantized

  #92891 merged May 7, 2025

• [XLA:GPU] Refactor RedzoneBuffers. Make pararameters of RedzoneBuffers explicit and decouple RedzoneBuffers from AutotunerCompileUtil and AutotunerCompileConfig.

  #92878 merged May 7, 2025

• [StableHLO] Fix typo.

  #92789 merged May 7, 2025

• [xla::ifrt] Rename xla::ifrt::Compiler::Compile to xla::ifrt::Compiler::CompileAndLoad and point xla::ifrt::Compiler::Compile to xla::ifrt::Compiler::CompileAndLoad.

  #92506 merged May 7, 2025

• Reverts eb4629ef35970666a34683d13d7fc8c5b421118a

  #92883 merged May 7, 2025

• #sdy Apply xla.sdy.FuncResultSharding if it has no uses or it's used by an op that isn't func.return.

  #92882 merged May 7, 2025

• [XLA:GPU] Simplify gpu_hlo_schedule_test using absl::ScopedMockLog.

  #92787 merged May 7, 2025

• Remove obsolete TODO

  #92799 merged May 7, 2025

• [XLA:GPU] Favor pipelined over synchronous collectives when combining.

  #92795 merged May 7, 2025

• [XLA:GPU] (NFC) Refactor ShapeToMlirTypes to use ForEachLeafShape.

  #92784 merged May 7, 2025

• [xla:cpu] Remove Tfrt from PJRT CPU buffer type names

  #92758 merged May 7, 2025

• [XLA:GPU] (NFC) Simplify logic in CreateMLIRModule.

  #92783 merged May 7, 2025

• PR #25919: feat: add runtime_version field to target_config

  #92711 merged May 7, 2025

• Record manual axes instead of generating mhlo.sharding in stablehlo/shard_map_export.

  #92533 merged May 7, 2025

• Automated Code Change

  #92863 merged May 7, 2025

• Automated Code Change

  #92857 merged May 7, 2025

• [XLA:GPU][Emitters] Correctly handle allocate_shared for s4 type.

  #92879 merged May 7, 2025

• Automated Code Change

  #92860 merged May 7, 2025

• [XLA:CPU] Migrate xla_jit_compiled_cpu_function to thunk execution

  #90379 merged May 7, 2025

• PR #25441: [XLA:GPU] add cuDNN Paged attention support

  #92703 merged May 7, 2025

• #sdy no need to add (and remove) a fake result to a callback custom call with no results.

  #92794 merged May 7, 2025

• PR #26101: Allow bitcasts in dynamic memcpy fusions.

  #92785 merged May 7, 2025

• PR #26111: [NFC] Refactor ResolveFunctionalDependencyOnInductionVariable.

  #92866 merged May 7, 2025

• [XLA:GPU] Add simple triton support test for tuple ops

  #91496 merged May 7, 2025

• [XLA:CPU] Expose interface for rendering execution graphs via dot API

  #92618 merged May 7, 2025

• Rollback oneDNN v3.7 + ACL v24.12 update in aarch64 build.

  #92809 merged May 7, 2025

• Use shared_ptr instead of unique_ptr for executable references

  #92826 merged May 7, 2025

• PR #25959: [ROCm] fixed missing headers + deps

  #92702 merged May 7, 2025

• [xla] Add a name to execution graph operations

  #92687 merged May 7, 2025

• Temporarily add UBDialect as a dependent dialect in SdyRoundTripStablehloToHloToStablehloPass.

  #92872 merged May 7, 2025

• [IFRT] Exclude sdy dialect from the verification conducted by the IFRT SPMD expandable verification pass

  #92770 merged May 7, 2025

• [XLA:GPU] Integrate perf tables to SoL estimator.

  #92412 merged May 7, 2025

• Bump the Shardy XLA pin.

  #92869 merged May 7, 2025

• Introduce a type alias for array references

  #92800 merged May 7, 2025

• Automated Code Change

  #92862 merged May 7, 2025

• fix(WeakTrackableReference): correct weak reference type handling

  #92336 merged May 7, 2025

• Temporarily roll back changes for new LLVM version

  #92828 merged May 7, 2025

• Added support for callback with LayoutModesToXlaShapes when auto-sharding modifies shapes of arguments or results.

  #92763 merged May 7, 2025

• Fork remaining passes to tf_passes and related .mlir tests in with new namespace

  #92541 merged May 7, 2025

• Use LoadedExecutableRef instead of std::unique_ptr<LoadedExecutable>

  #92815 merged May 6, 2025

• Replace usage of scipy.misc.derivative in TensorFlow.

  #92822 merged May 6, 2025

• Add two optimization patterns replacing x + (y - y) or (y - y) + x with x.

  #92811 merged May 6, 2025

• Remove tf/compiler/mlir/quantization/common/quantization_lib and tf/compiler/mlir/quantization/tensorflow/ops:tf_op_quant_spec from tf_passes

  #92730 merged May 6, 2025

• Integrate StableHLO at openxla/stablehlo@63246aea

  #92446 merged May 6, 2025

• [XLA] Remove UB dialect from XLA dependencies.

  #92722 merged May 6, 2025

• [xla:gpu] Detect empty traced CUDA graphs

  #92806 merged May 6, 2025

• Add Fprint128ToBytes utility function.

  #92749 merged May 6, 2025

• Reverts 38a0e6bdaea46322fbeffb0902a8136fe8bc9130

  #92803 merged May 6, 2025

• Remove tf/compiler/mlir/lite:tensorflow_lite from tf/compiler/mlir/quantization/tensorflow:passes

  #92416 merged May 6, 2025

• Move BufferAllocationSliceProto to xla/service/buffer_assignment.proto

  #92786 merged May 6, 2025

• Precompute set of buffers allocated/released per instruction.

  #92739 merged May 6, 2025

• Add swizzle_mode to TMA attribute

  #92790 merged May 6, 2025

• [xla:ffi] Fix typo in struct size used during DeviceOrdinal decoding.

  #92797 merged May 6, 2025

• PR #26100: Clean up special cases for AsyncStart in HloComputation.

  #92780 merged May 6, 2025

• Add a bazel config setting to indicate whether the code is built in a CI.

  #92516 merged May 6, 2025

• Introduce aliases for executable references

  #92792 merged May 6, 2025

• PR #22788: Implement GetCompiledMemoryStats for GPU AOT executables

  #92796 merged May 6, 2025

• Integrate support for tensordesc structs and reenable descriptor tests

  #92411 merged May 6, 2025

• Add proto serialization for the SequentialThunk

  #92772 merged May 6, 2025

• [XLA:GPU] Add CuBLAS autotuner backend.

  #92269 merged May 6, 2025

• Rewrite the argument filtering code to use argument masks

  #92410 merged May 6, 2025

• [XLA:GPU]: Remove GpuPerformanceModelOptions

  #92717 merged May 6, 2025

• #sdy Handle the case where an output of a sdy.manual_computation has a dimension of size zero.

  #92781 merged May 6, 2025

• Remove unused dependency on flatbuffer_import

  #92728 merged May 6, 2025

• [XLA:GPU] Add support for .txtpb files.

  #92743 merged May 6, 2025

• [XLA:GPU] Remove unused trace argument from CreateMLIRModule.

  #92779 merged May 6, 2025

• Drop unneeded dep from dialect

  #92606 merged May 6, 2025

• PR #26046: Generalize dynamic memcpy support.

  #92713 merged May 6, 2025

• Automated Code Change

  #92691 merged May 6, 2025

• [xla:gpu][triton] Update TMA lowering in XLA to use new api

  #92716 merged May 6, 2025

• [XLA:GPU] Collect matmul perf tables on forge.

  #92712 merged May 6, 2025

• [XLA] Disable dynamic-slice async conversion until we correctly account for operands.

  #92757 merged May 6, 2025

• Automated Code Change

  #92647 merged May 6, 2025

• Remove unused header and dep.

  #92707 merged May 6, 2025

• [XLA:GPU] Remove xla_gpu_unsafe_pipelined_loop_annotator flag

  #89920 merged May 6, 2025

• [XLA:GPU] Remove special case for send/recv in async events mapping

  #92740 merged May 6, 2025

• Fix missing dep.

  #92727 merged May 6, 2025

• [XLA:GPU] Remove brittle pipeline parallelism detection in the SPMD partitioner pass

  #90065 merged May 6, 2025

• Support platform / memory space allocator in TfrtGpuClient.

  #92750 merged May 6, 2025

• [xla:pjrt] Add micro benchmarks for StreamExecutorGpuClient

  #92738 merged May 6, 2025

• Initialize fp16_to_fp32 with identity map

  #92633 merged May 6, 2025

• [XLA:GPU] Move all_reduce_splitter to collectives folder

  #92511 merged May 6, 2025

• TransposeConv bias should be matching the dimension of the output channel, and should always be 1D if exists. Flat the succeeding bias constant before fusion.

  #92689 merged May 6, 2025

• Add tfl.EmbeddingLookup DRQ

  #92742 merged May 6, 2025

• [XLA:GPU] Move all_reduce_decomposer to collectives folder

  #92512 merged May 6, 2025

• [XLA:GPU] Move all_reduce_blueconnect to collectives folder

  #92509 merged May 5, 2025

• Updates IOPDDL conversion to shift edge costs by a constant amount if negative values are present.

  #92726 merged May 5, 2025

• Implement TransferToInfeed and TransferFromOutfeed in pjrt async gpu client.

  #92588 merged May 5, 2025

• Internal only changes

  #92733 merged May 5, 2025

• [XLA:GPU] Move all_gather_dynamic_slice_simplifier to collectives folder

  #92510 merged May 5, 2025

• Automated Code Change

  #92719 merged May 5, 2025

• [XLA:benchmarks] Update benchmark proto and add implementations to load and parse proto for triggering benchmark run using GitHub Actions

  #92619 merged May 5, 2025

• [XLA:GPU] Move all_gather_optimizer to collectives folder

  #92508 merged May 5, 2025

• Refactor HloRunnerPjRt to distinguish loaded and unloaded PjRtExecutables.

  #92630 merged May 5, 2025

• [IFRT] Switch std::shared_ptr<xla::ifrt::Sharding> to xla::ifrt::ShardingRef

  #92627 merged May 5, 2025

• Prepare VarHandle ops too

  #92721 merged May 5, 2025

• Avoid host scheduling dependencies between two GPU CUDA requests

  #91890 merged May 5, 2025

• [JAX] In McJAX, support JIT compilation/execution on only the devices attached to a single host, with other hosts not participating.

  #91849 merged May 5, 2025

• Implement Auotuner Backend interface for Triton.

  #92710 merged May 5, 2025

• [XLA:GPU/TMA] Modified some TMA tests based on investigations in their failures. Incorrect results were caused by incorrect handling of swizzling modes with regards to TMA. After the latest Triton integration, swizzling is determined during the compilation pipeline and should be extracted from TTGIR. This will be addressed independently.

  #92718 merged May 5, 2025

• [XLA:GPU] update autotune db entry

  #92715 merged May 5, 2025

• [XLA:GPU] Add simple triton support test for map op

  #91655 merged May 5, 2025

• Automated Code Change

  #92645 merged May 5, 2025

• Remove more if_gpu_is_configured usages from service/gpu/BUILD

  #92486 merged May 5, 2025

• [XLA:GPU] Check whether the propagated tile offsets can be used.

  #90805 merged May 5, 2025

• PR #25953: [XLA:FFI][BugFix] Add Token Args and Rets for typed FFI on CPU backend

  #92709 merged May 5, 2025

• PR #25916: Added SafeStaticInit for deadlock free safe static initialization

  #92708 merged May 5, 2025

• PR #25965: [NVIDIA GPU] Fix fabric info test

  #92607 merged May 5, 2025

• PR #26003: Run fewer analyses in HloEvaluator.

  #92616 merged May 5, 2025

• PR #26001: [NVIDIA GPU] Change UB register cache to include size in the key

  #92705 merged May 5, 2025

• PR #24279: Handle FP8 bitcasting when lowering tensors

  #92611 merged May 5, 2025

• PR #26012: Convert computations to MakeNonfusionComputations in async wrappers

  #92704 merged May 5, 2025

• [XLA:Python] [JAX] Move HloPass bindings to XLA/Python.

  #92688 merged May 5, 2025

• Automated Code Change

  #92676 merged May 4, 2025

228 Pull requests opened by 7 people

• Get rid of `HloModuleProto` from `CompiledMemoryStats`

  #92690 opened May 4, 2025

• Automated Code Change

  #92692 opened May 5, 2025

• Automated Code Change

  #92693 opened May 5, 2025

• Automated Code Change

  #92694 opened May 5, 2025

• Automated Code Change

  #92695 opened May 5, 2025

• Automated Code Change

  #92696 opened May 5, 2025

• Automated Code Change

  #92697 opened May 5, 2025

• Automated Code Change

  #92698 opened May 5, 2025

• Automated Code Change

  #92699 opened May 5, 2025

• Automated Code Change

  #92700 opened May 5, 2025

• Automated Code Change

  #92701 opened May 5, 2025

• TfLite elementwise_ops add type support (#104)

  #92706 opened May 5, 2025

• TfLite exp missing datatype support (#69) (#121)

  #92714 opened May 5, 2025

• Explicitly include `errors.h` in `status_macros.h`. We need `tsl::error::INTERNAL` from it.

  #92723 opened May 5, 2025

• bug fix in literal comparison in `equal` test.

  #92724 opened May 5, 2025

• remove transitive lite dependencies from tf_passes

  #92731 opened May 5, 2025

• [xla::PyClient] Update PyClient to use xla::ifrt::CompileAndLoad.

  #92732 opened May 5, 2025

• PR #25635: Add parsing debug options support

  #92734 opened May 5, 2025

• [XLA:GPU] Return an error if `GetSchedulerMemoryLimit` fails.

  #92735 opened May 5, 2025

• Collect execution time for mlrt.

  #92744 opened May 5, 2025

• [Cleanup] Use CHECK_NOTNULL and CHECK_OK

  #92745 opened May 5, 2025

• Fix clang-format and BUILD dependency issues

  #92751 opened May 5, 2025

• Update GetLiveNodes API to be a bit more robust to worker restarts.

  #92752 opened May 5, 2025

• Integrate LLVM at llvm/llvm-project@0009a1783490

  #92754 opened May 5, 2025

• Automated Code Change

  #92756 opened May 5, 2025

• [JAX] Make fully replicated sharding to avoid materializing the same host buffers

  #92759 opened May 6, 2025

• Automated Code Change

  #92760 opened May 6, 2025

• Automated Code Change

  #92761 opened May 6, 2025

• Automated Code Change

  #92762 opened May 6, 2025

• Automated Code Change

  #92764 opened May 6, 2025

• Automated Code Change

  #92765 opened May 6, 2025

• Fix missing dep.

  #92766 opened May 6, 2025

• Automated Code Change

  #92767 opened May 6, 2025

• Delete constructor for tf_allocator_adapter.

  #92768 opened May 6, 2025

• Automated Code Change

  #92773 opened May 6, 2025

• Set wheel CUDA versions to be minor-version lower bound.

  #92774 opened May 6, 2025

• Automated Code Change

  #92775 opened May 6, 2025

• [XLA:GPU][Emitters] Use new heuristic for transpose emitter

  #92776 opened May 6, 2025

• PR #26100: Clean up special cases for AsyncStart in HloComputation.

  #92778 opened May 6, 2025

• [xla:gpu] Fork existing fusion_emitter_int4_device_test.cc and make it use generic emitter.

  #92801 opened May 6, 2025

• Add [de]serialization for `GemmConfig`.

  #92802 opened May 6, 2025

• Set remaining profile options and perform validation on the advanced_configuration keys.

  #92804 opened May 6, 2025

• Add [de]serialization for `BufferAllocationSlice`.

  #92805 opened May 6, 2025

• [XLA:Collective] Cache the results from GetParticipatingFlattenedIdGroups().

  #92813 opened May 6, 2025

• Add specialization for `PrintToStream` in `TF_RET_CHECK`'s additional output.

  #92814 opened May 6, 2025

• Reverts 6932184196b02be51a5be4ca1090a422f897dc4e

  #92817 opened May 6, 2025

• Integrate LLVM at llvm/llvm-project@3e235a7c601d

  #92818 opened May 6, 2025

• Make Shape reference-counted with copy-on-write.

  #92819 opened May 6, 2025

• Replace `std::shared_ptr<xla::ifrt::LoadedExecutable>` with `xla::ifrt::LoadedExecutableRef`

  #92827 opened May 7, 2025

• Prevent MSA from creating a worst case unbalanced buffer interval tree when allocating scoped vmem.

  #92829 opened May 7, 2025

• Use utility functions to check number of inputs and outputs.

  #92832 opened May 7, 2025

• Automated Code Change

  #92833 opened May 7, 2025

• Automated Code Change

  #92834 opened May 7, 2025

• Automated Code Change

  #92835 opened May 7, 2025

• Automated Code Change

  #92836 opened May 7, 2025

• Automated Code Change

  #92837 opened May 7, 2025

• Automated Code Change

  #92838 opened May 7, 2025

• Automated Code Change

  #92839 opened May 7, 2025

• Automated Code Change

  #92840 opened May 7, 2025

• Automated Code Change

  #92841 opened May 7, 2025

• Automated Code Change

  #92842 opened May 7, 2025

• Automated Code Change

  #92843 opened May 7, 2025

• Automated Code Change

  #92844 opened May 7, 2025

• Automated Code Change

  #92845 opened May 7, 2025

• Automated Code Change

  #92846 opened May 7, 2025

• Automated Code Change

  #92847 opened May 7, 2025

• Automated Code Change

  #92848 opened May 7, 2025

• Automated Code Change

  #92849 opened May 7, 2025

• Automated Code Change

  #92850 opened May 7, 2025

• Automated Code Change

  #92851 opened May 7, 2025

• Automated Code Change

  #92852 opened May 7, 2025

• Automated Code Change

  #92853 opened May 7, 2025

• Automated Code Change

  #92854 opened May 7, 2025

• Automated Code Change

  #92855 opened May 7, 2025

• [XLA:GPU] Add simple triton support test for topK op

  #92856 opened May 7, 2025

• Automated Code Change

  #92858 opened May 7, 2025

• Automated Code Change

  #92859 opened May 7, 2025

• Automated Code Change

  #92861 opened May 7, 2025

• Automated Code Change

  #92864 opened May 7, 2025

• Automated Code Change

  #92865 opened May 7, 2025

• [xla:gpu] add extract_tma_info pass

  #92867 opened May 7, 2025

• Update TmaDescriptor creation to use triton's internal swizzling decision

  #92868 opened May 7, 2025

• Automated Code Change

  #92870 opened May 7, 2025

• Automated Code Change

  #92871 opened May 7, 2025

• Reverts f6269e13f1d709b4fe46079e36f1407f65075271

  #92874 opened May 7, 2025

• Reenable tma_to_llvm.mlir test

  #92875 opened May 7, 2025

• feat: add datatype support for add, ceil, mul, range, sign, sub

  #92880 opened May 7, 2025

• #sdy Cleanup `removeAutoAxesToAvoidPadding` as this is now done in `-sdy-update-non-divisible-input-output-shardings`.

  #92885 opened May 7, 2025

• [xla:ifrt] Rename references to ::Compile(module, topology, options) that return LoadedExecutables to ::CompileAndLoad.

  #92892 opened May 7, 2025

• [XLA:CPU][roll-forward] Internal fixes.

  #92894 opened May 7, 2025

• fork quantize_passes to use tf_passes

  #92895 opened May 7, 2025

• Integrate LLVM at llvm/llvm-project@ed8d4fe0b1d8

  #92896 opened May 7, 2025

• Custom NVPTX backend.

  #92900 opened May 7, 2025

• Integrate LLVM at llvm/llvm-project@2d287f51eff2

  #92901 opened May 7, 2025

• Introduce the `BUFFER` primitive type and update the `Shape` class to support it.

  #92906 opened May 7, 2025

• [XLA:benchmarks] Add workflow files to generate GHA input matrices and a presubmit workflow

  #92912 opened May 7, 2025

• Add traceme while borrowing streams

  #92917 opened May 7, 2025

• Implement default `layout` in IfrtProxyArray

  #92923 opened May 8, 2025

• Automated Code Change

  #92925 opened May 8, 2025

• Automated Code Change

  #92926 opened May 8, 2025

• Automated Code Change

  #92927 opened May 8, 2025

• Automated Code Change

  #92928 opened May 8, 2025

• Automated Code Change

  #92929 opened May 8, 2025

• Automated Code Change

  #92930 opened May 8, 2025

• Automated Code Change

  #92932 opened May 8, 2025

• Automated Code Change

  #92933 opened May 8, 2025

• Automated Code Change

  #92934 opened May 8, 2025

• Automated Code Change

  #92935 opened May 8, 2025

• Automated Code Change

  #92936 opened May 8, 2025

• Automated Code Change

  #92937 opened May 8, 2025

• Automated Code Change

  #92947 opened May 8, 2025

• Automated Code Change

  #92949 opened May 8, 2025

• Increase random UUID space in the fingerprint proto.

  #92952 opened May 8, 2025

• Automated Code Change

  #92954 opened May 8, 2025

• Handle non-normalized layouts for TMA.

  #92957 opened May 8, 2025

• [xla:gpu] Use enum for swizzle mode in tma_descriptor attributes.

  #92958 opened May 8, 2025

• [XLA:GPU] Add option to always fold constants

  #92962 opened May 8, 2025

• fork stablehlo-quant-opt to tf-stablehlo-quant-opt without lite deps

  #92972 opened May 8, 2025

• [XLA:GPU] Add triton support test for triangular-solve op

  #92973 opened May 8, 2025

• [ifrt] Refactor away from deprecated constructors

  #92978 opened May 8, 2025

• Replace calls to <ctype.h> functions with their absl:: equivalents.

  #92981 opened May 8, 2025

• Convert remaining `native.py_*` usages to load from rules_python

  #92984 opened May 8, 2025

• Remove TfrtCpuClient::BufferFromHostLiteral in favor of CommonPjRtClient::BufferFromHostLiteral.

  #92985 opened May 8, 2025

• Make HloInstruction::name_ only take 8 bytes inline in the object

  #92986 opened May 8, 2025

• [tosa] Fix `obj.isa` -> `mlir::isa` in an assertion

  #92987 opened May 8, 2025

• Consistently apply preprocessing in HloRunnerAgnosticTestBase + friends.

  #92988 opened May 8, 2025

• Integrate LLVM at llvm/llvm-project@741fef3a4453

  #92993 opened May 9, 2025

• Move CreateViewOfDeviceBuffer and BufferFromHostBuffer into CommonPjRtClient

  #92994 opened May 9, 2025

• Remove support for dimension level type.

  #92995 opened May 9, 2025

• Automated Code Change

  #92996 opened May 9, 2025

• Automated Code Change

  #92997 opened May 9, 2025

• Automated Code Change

  #92998 opened May 9, 2025

• Automated Code Change

  #92999 opened May 9, 2025

• Automated Code Change

  #93000 opened May 9, 2025

• Automated Code Change

  #93002 opened May 9, 2025

• Automated Code Change

  #93004 opened May 9, 2025

• Automated Code Change

  #93005 opened May 9, 2025

• Automated Code Change

  #93006 opened May 9, 2025

• Automated Code Change

  #93007 opened May 9, 2025

• Automated Code Change

  #93009 opened May 9, 2025

• Make the recently-added RegisterPluggableDevicePlugin type-safe.

  #93010 opened May 9, 2025

• PR #26255: [Nvidia GPU] Nvshmem communicator API

  #93011 opened May 9, 2025

• Remove ConvertFakeQuantWithMinMaxVarsOp from TFLite lowering patterns.

  #93012 opened May 9, 2025

• Automated Code Change

  #93013 opened May 9, 2025

• Automated Code Change

  #93014 opened May 9, 2025

• PR #25601: Fix NCCL user buffer allocation error message and unit tests for kCollective

  #93015 opened May 9, 2025

• [XLA:CPU] Graph visualization called only for executor owned by CpuExecutable

  #93016 opened May 9, 2025

• PR #26255: [Nvidia GPU] Nvshmem communicator API

  #93017 opened May 9, 2025

• Integrate Triton up to [fe66e412](https://github.com/openai/triton/commits/fe66e41296ec0fad9058792714ffa146bdf9253a)

  #93018 opened May 9, 2025

• [XLA:CPU][XLA:GPU] Implement Shared `ForAll` op and use it in GPU loop fusion

  #93020 opened May 9, 2025

• [XLA:CPU] Implement lowering of shared `forall` op

  #93021 opened May 9, 2025

• [XLA:GPU] Add triton support test for convolution op

  #93022 opened May 9, 2025

• [XLA:CPU] Add op type id to ExecutionGraph::Operation

  #93023 opened May 9, 2025

• [XLA][graphviz] Graph visualization from a span of ops

  #93026 opened May 9, 2025

• [XLA:CPU][XLA:GPU] Implement shared workgroup id op

  #93027 opened May 9, 2025

• [XLA] Move execution graph renderer to execution_graph

  #93028 opened May 9, 2025

• internal BUILD rule visibility

  #93031 opened May 9, 2025

• Automated Code Change

  #93032 opened May 9, 2025

• [XLA:GPU] Add custom kernel autotuning backend.

  #93034 opened May 9, 2025

• [XLA:GPU] Only support cuBLAS custom calls in cuBLAS autotuning backend.

  #93036 opened May 9, 2025

• Manually migrate deprecated references to `strings::StrCat` and `strings::StrAppend`.

  #93039 opened May 9, 2025

• Add proto serialization for ShapeIndex

  #93040 opened May 9, 2025

• Push down the logic that reads the default memory fraction for GPUs from `XLA_CLIENT_MEM_FRACTION`

  #93041 opened May 9, 2025

• [XLA:CPU] Add lowering for `workgroup_id` to `thread_id`

  #93042 opened May 9, 2025

• [XLA:CPU] Add ability to get non-x thread dimension in cpu fusion dialect.

  #93043 opened May 9, 2025

• update sharding settings to remove empty shards in execution

  #93045 opened May 9, 2025

• Use the shell rules https://github.com/bazelbuild/rules_shell.

  #93047 opened May 9, 2025

• #sdy save shardings of the arguments on `CaseOp` during MHLO->HLO translation.

  #93048 opened May 9, 2025

• Use low latency thread pool for async PjRT

  #93052 opened May 9, 2025

• Don't print a warning when a dependency repo is already defined.

  #93055 opened May 9, 2025

• [XLA:CPU] Expose an interface to get nested thunks

  #93056 opened May 9, 2025

• [XLA][graphviz] Visualize nested thunks as clusters

  #93057 opened May 9, 2025

• Fix cost analysis on for output byte accessed when result is tuple

  #93059 opened May 9, 2025

• Allow simplify_ici_dummy_variables to be triggered with TPUExecuteAndUpdateVariables nodes.

  #93060 opened May 9, 2025

• Upgrade bazel_skylib to 1.7.1

  #93061 opened May 9, 2025

• Log before and after calling test subprocess.

  #93062 opened May 9, 2025

• Internal visibility changes

  #93064 opened May 9, 2025

• Remove the semaphore in ExecuteHelper

  #93066 opened May 9, 2025

• Automated Code Change

  #93069 opened May 10, 2025

• Automated Code Change

  #93070 opened May 10, 2025

• Automated Code Change

  #93073 opened May 10, 2025

• Automated Code Change

  #93074 opened May 10, 2025

• Automated Code Change

  #93075 opened May 10, 2025

• Automated Code Change

  #93076 opened May 10, 2025

• Automated Code Change

  #93077 opened May 10, 2025

• Automated Code Change

  #93078 opened May 10, 2025

• Cache invalidation for XProf

  #93080 opened May 10, 2025

• Automated Code Change

  #93081 opened May 10, 2025

• Automated Code Change

  #93082 opened May 10, 2025

• Automated Code Change

  #93083 opened May 10, 2025

• Automated Code Change

  #93084 opened May 10, 2025

• Automated Code Change

  #93085 opened May 10, 2025

• Automated Code Change

  #93086 opened May 10, 2025

• Automated Code Change

  #93087 opened May 10, 2025

• Automated Code Change

  #93088 opened May 10, 2025

• Automated Code Change

  #93090 opened May 10, 2025

• Automated Code Change

  #93091 opened May 10, 2025

• Automated Code Change

  #93093 opened May 10, 2025

• Update activations.py

  #93094 opened May 10, 2025

• [XLA] Simplify CheckInplaceCollectivePermute.

  #93096 opened May 10, 2025

• Reverts fdde76d61ee110eb0e1dad27e9af5589bda9ef11

  #93097 opened May 10, 2025

• [XLA] Remove ShapeUtil::GetSubshapeOneIndex().

  #93098 opened May 10, 2025

• Use a tile assignment to store device groups in grouped sharding.

  #93103 opened May 10, 2025

• Fix a bug where TFRT GPU does not handle token literals correctly

  #93104 opened May 10, 2025

• Automated Code Change

  #93106 opened May 11, 2025

• Automated Code Change

  #93107 opened May 11, 2025

• Automated Code Change

  #93108 opened May 11, 2025

• Automated Code Change

  #93109 opened May 11, 2025

• Automated Code Change

  #93110 opened May 11, 2025

• Automated Code Change

  #93111 opened May 11, 2025

• Automated Code Change

  #93112 opened May 11, 2025

• Automated Code Change

  #93113 opened May 11, 2025

• Automated Code Change

  #93114 opened May 11, 2025

• Automated Code Change

  #93115 opened May 11, 2025

• Automated Code Change

  #93116 opened May 11, 2025

• Automated Code Change

  #93117 opened May 11, 2025

• Automated Code Change

  #93118 opened May 11, 2025

• Automated Code Change

  #93119 opened May 11, 2025

• Automated Code Change

  #93120 opened May 11, 2025

• Automated Code Change

  #93121 opened May 11, 2025

• Automated Code Change

  #93122 opened May 11, 2025

• Automated Code Change

  #93123 opened May 11, 2025

• Remove Tfrt from PJRT GPU buffer type names

  #93124 opened May 11, 2025

• Add complex64 and complex128 support to tf.truncatediv

  #93125 opened May 11, 2025

• Automated Code Change

  #93126 opened May 11, 2025

• Automated Code Change

  #93127 opened May 11, 2025

• Automated Code Change

  #93128 opened May 11, 2025

• Automated Code Change

  #93129 opened May 11, 2025

17 Issues closed by 5 people

• The pip instructions to download and install tensorflow for macOS don't work

  #92062 closed May 11, 2025

• tensorflow dll issue

  #93008 closed May 10, 2025

• DLL load failed while importing _pywrap_tensorflow_internal

  #92941 closed May 10, 2025

• Grateful for Your Work — Open to Collaborate

  #93092 closed May 10, 2025

• `tf.math.imag` doesn't support float and double

  #92876 closed May 10, 2025

• warning in official

  #91733 closed May 9, 2025

• Compiling TFLite introductory example fails due to missing headers

  #91782 closed May 9, 2025

• check

  #92920 closed May 8, 2025

• Install TensorFlow with pip for macOS doesn't work on M4 MacBook Pro Apple Silicon

  #91681 closed May 7, 2025

• Tensorflow model.save fails with AttributeError: 'Variable' object has no attribute 'read_value'

  #91686 closed May 7, 2025

• Loading tensorflow DLL error in Spyder other python ok

  #91709 closed May 7, 2025

• Mac Air M2, TfLiteGpuDelegate, Cpp, C++ Building from source

  #86075 closed May 6, 2025

• CMake Error: install(EXPORT "tensorflow-liteTargets" ...) includes target "tensorflow-lite" which requires target "pthreadpool" that is not in any export set.

  #87172 closed May 6, 2025

• FP32 gives good results but int8 predicts only single class

  #89438 closed May 6, 2025

• Tensorflow Build Error

  #91558 closed May 6, 2025

• multiprocessing incompatibility

  #91647 closed May 6, 2025

• MemoryError during "import tensorflow" with tensorflow-cpu==2.19.0

  #89268 closed May 5, 2025

10 Issues opened by 7 people

• `../tensorflow/third_party/xla/third_party/tsl/tsl/platform/ml_dtypes.h:19:10: error: 'ml_dtypes/include/float8.h' file not found [clang-diagnostic-error]`

  #93130 opened May 11, 2025

• clang: error: unknown argument: '-fno-canonical-system-headers'

  #93049 opened May 9, 2025

• Inconsistencies between the implementation and documentation of `tf.strided_slice`

  #92955 opened May 8, 2025

• Build TensorFlow Lite C library Error: no such instruction: 'vdpbf16ps zmm7, zmm2, zmm10'

  #92945 opened May 8, 2025

• [iOS 18] EXC_BAD_ACCESS when initializing TFLite model on device running iOS 18.4.1

  #92924 opened May 8, 2025

• `tf.truncatediv` doesn't work on complex64 and complex128

  #92922 opened May 8, 2025

• `tf.math.reciprocal` doesn't work on int8, int16 and int32

  #92921 opened May 8, 2025

• `tf.math.igamma` doesn't work when dtype is bfloat16 or half

  #92873 opened May 7, 2025

• Compatibility Challenges of Transformer Models (MobileBERT, Mobile ViT) with TensorFlow Lite and GPU Delegation

  #92798 opened May 6, 2025

• Different inference results on CPU vs GPU for simple model on macOS M1

  #92686 opened May 4, 2025

59 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.

• Problem when using tf-to-tosa-pipeline

  #92660 commented on May 5, 2025 • 0 new comments

• Error when converting tensorflow model in mlir

  #92659 commented on May 5, 2025 • 0 new comments

• GPU kernel for `tf.linalg.eig` / `tf.linalg.eigvals`

  #92194 commented on May 5, 2025 • 0 new comments

• New version of nVidia Windows GPU driver (576+) causing TF in WSL process to crash.

  #92251 commented on May 5, 2025 • 0 new comments

• TensorFlow on RTX 5090

  #89272 commented on May 5, 2025 • 0 new comments

• The default value for skip_gradients_aggregation (False) of optimizers prevents freeing memory

  #92094 commented on May 5, 2025 • 0 new comments

• Impossible to free GPU memory used by rank 0 tensors

  #92103 commented on May 5, 2025 • 0 new comments

• How to setup tensorflow environment in google colab for training ai models using tensorflow object detection APIs?

  #91738 commented on May 6, 2025 • 0 new comments

• [TFLite] GPU delegation issues with Vision Transformer (ViT-base) in TFLite: dynamic shape & op support problems

  #92098 commented on May 6, 2025 • 0 new comments

• Muting Tensorflow Lite logs

  #92216 commented on May 6, 2025 • 0 new comments

• model load problem

  #92044 commented on May 7, 2025 • 0 new comments

• Compiling Tensorflow v2.19.0 with mkl_aarch64 fails.

  #92279 commented on May 7, 2025 • 0 new comments

• GPU Illegal memory access in kernel DenseToCSRSparseMatrix

  #92413 commented on May 7, 2025 • 0 new comments

• Tensorflow for Arch Linux

  #92550 commented on May 7, 2025 • 0 new comments

• No way to use tensorflow with cuda on windows. cudaGetDevice() failed.

  #92523 commented on May 10, 2025 • 0 new comments

• "Given shapes (...) are not broadcastable" after conversion onnx->tf->tflite

  #92037 commented on May 10, 2025 • 0 new comments

• Setting the protobuf version at installation with bazel

  #92610 commented on May 10, 2025 • 0 new comments

• It doesn't support on python3.13

  #78774 commented on May 11, 2025 • 0 new comments

• Add ComplexOp as a builtin to TFLite

  #61359 commented on May 5, 2025 • 0 new comments

• Support for arbitrary tensor sizes in `tfl.strided_slice` lowering

  #61424 commented on May 5, 2025 • 0 new comments

• Build kernel for all numbertypes for ApproximateEqual Op

  #64587 commented on May 5, 2025 • 0 new comments

• Fix checkfail in MatrixSetDiagV3

  #65702 commented on May 5, 2025 • 0 new comments

• iwyu: missing includes in xla

  #66450 commented on May 5, 2025 • 0 new comments

• Add GPU configuration options to TensorFlow Go bindings

  #81989 commented on May 7, 2025 • 0 new comments

• Remove Transfer-Encoding: identity considered to be illegal

  #83082 commented on May 5, 2025 • 0 new comments

• [oneDNN][CPU] fuse a matmul pattern

  #86172 commented on May 5, 2025 • 0 new comments

• fix(kernels): Handle empty values with non-empty row splits in RaggedTensorToTensor

  #86349 commented on May 5, 2025 • 0 new comments

• Fix compile error in tensorflow/python/tfcompile_wrapper.cc on s390x

  #87676 commented on May 5, 2025 • 0 new comments

• TfLite. Fix of issue 61269

  #87946 commented on May 10, 2025 • 0 new comments

• TfLite. Fix of issue 54269

  #88020 commented on May 11, 2025 • 0 new comments

• [XLA:CPU][tfcompile] Use thunk runtime for tfcompiled models.

  #88552 commented on May 6, 2025 • 0 new comments

• Fix: Ensure boolean_mask_v2() only accepts boolean dtype for mask

  #89370 commented on May 5, 2025 • 0 new comments

• Move duplicate CUDA/XLA registration logs from INFO to VLOG

  #89808 commented on May 5, 2025 • 0 new comments

• Adapt commit 5467ee9 from XLA to TensorFlow

  #90563 commented on May 6, 2025 • 0 new comments

• Create a PjRt protos dir.

  #90719 commented on May 9, 2025 • 0 new comments

• adam opt .

  #90795 commented on May 5, 2025 • 0 new comments

• #sdy Properly handle token types in JAX and `ManualComputationOp`.

  #91048 commented on May 7, 2025 • 0 new comments

• Enable Android XNNPACK delegate num_threads with Interpreter.Options.…

  #91181 commented on May 6, 2025 • 0 new comments

• experimental test for dup registration bug

  #91758 commented on May 7, 2025 • 0 new comments

• Major deps update:

  #92241 commented on May 11, 2025 • 0 new comments

• [XLA:ReplicaGroupV2] Propagate ReplicaGroupV2 to more places. (NFC)

  #92319 commented on May 7, 2025 • 0 new comments

• [XLA:GPU][Emitters] Scan through the IR to compute shared memory usage.

  #92427 commented on May 5, 2025 • 0 new comments

• [XLA] Add F8E8M0FNU to IsF8Type

  #92441 commented on May 9, 2025 • 0 new comments

• feat: add datatype support for tfl.transpose

  #92472 commented on May 6, 2025 • 0 new comments

• Do not create BFCAllocator if not needed

  #92483 commented on May 5, 2025 • 0 new comments

• Update Riegeli dependency to the latest commit.

  #92502 commented on May 10, 2025 • 0 new comments

• Fixes #58106: Add shape validation to AvgPool3DGrad input to prevent crash

  #92521 commented on May 8, 2025 • 0 new comments

• Add benchmark for Shape assignment.

  #92548 commented on May 8, 2025 • 0 new comments

• Switch uses of `shape.DebugString()` to `shape.ToString()` or `shape.ToProto().DebugString()`.

  #92572 commented on May 5, 2025 • 0 new comments

• PR #23884: [NVIDIA] Fix a cast error in checking overlap limit function in LHS

  #92573 commented on May 9, 2025 • 0 new comments

• Composite,CustomCall,DotGeneral,Dot,DynamicConv,DynamicGather,DynamicIota,DynamicPad,DynamicReshape,Reshape: Direct StableHLO to HLO Translation

  #92592 commented on May 6, 2025 • 0 new comments

• Automated Code Change

  #92597 commented on May 7, 2025 • 0 new comments

• Automated Code Change

  #92598 commented on May 7, 2025 • 0 new comments

• Automated Code Change

  #92600 commented on May 9, 2025 • 0 new comments

• Fix a bug in `RandomPathGreedy` when tracking the objective value over iterations; previously, it adds `cost_delta` to `current_cost` even if a new strategy is rejected due to the peak-memory violation. Also, streamline the previous CL by combining the previous two phases (Phase I and II).

  #92622 commented on May 8, 2025 • 0 new comments

• Remove the lite/toco protobuf files out of //tensorflow/core:lib_internal_impl

  #92624 commented on May 5, 2025 • 0 new comments

• Implement Hadamard rotation reference as a custom op

  #92635 commented on May 7, 2025 • 0 new comments

• Automated Code Change

  #92649 commented on May 6, 2025 • 0 new comments

• Automated Code Change

  #92681 commented on May 7, 2025 • 0 new comments