Repository navigation

[pull] master from tensorflow:master - #253

Open

pull[bot] wants to merge 10000 commits into

greedforgood:masterfrom

tensorflow:master

Open

[pull] master from tensorflow:master#253
pull[bot] wants to merge 10000 commits into
greedforgood:masterfrom
tensorflow:master

This pull request is big! We're only showing the most recent 250 commits

Commits on Oct 14, 2026
- PR #50654: [XLA:GPU] Set fusion kind to kCustom when applying a cuDNN config
  
  loupicaaa
  authored and
  tensorflower-gardener
  committed
  1b4aaea
- [XLA:GPU] Extract GetCollectiveInfo helper function into utils.
  
  olegshyshkov
  authored and
  tensorflower-gardener
  committed
  5d2cb43
- Revert rank-check test regression and guard scalar elements
  MaddipatlaChetan24
  committed
  154c9c3
- Fix test_checkpoint_deferred_restore_scalar_dvariable CI failure
  MaddipatlaChetan24
  committed
  efd5c92
- Add `BarrierMode::kPeerSymmetric` (`peer_symmetric`) to `triton_xla.block_barrier` for 1-to-1 pairwise synchronization between corresponding blocks across peer GPUs.
  
  PatriosTheGreat
  authored and
  tensorflower-gardener
  committed
  a774de5
- Restore config.py; apply empty/whitespace DTENSOR_JOBS fix correctly
  MaddipatlaChetan24
  committed
  1e2703e
- Fix PyLint errors: unused imports and stray Py2.7 compat shim
  MaddipatlaChetan24
  committed
  cf763d4
- Reverts fcb146d1780ac754ca435fca708c76fdad743cd5
  tensorflower-gardener
  committed
  6202f53
- PR #50613: [XLA:GPU][SPIR-V] Fix SPIR-V target lowering object file initialization
  
  akhilgoe
  authored and
  tensorflower-gardener
  committed
  15b56ae
- Fix unbracketed IPv6 parsing and context-mutation ordering
  MaddipatlaChetan24
  committed
  5f816c9
- Rename windows_lib_diagnostics test file and fix PATH test
  MaddipatlaChetan24
  committed
  e86e91f
- Add regression tests for throughput division-by-zero and filter regex fix
  MaddipatlaChetan24
  committed
  81ae34c
- Automated Code Change
  tensorflower-gardener
  committed
  c24e5ec
- Restore c_api_util_test.py and fix ScopedTFBuffer.__del__ for None c_api
  MaddipatlaChetan24
  committed
  ff64aa8
- [XLA:GPU] Pass k_loop_iterations directly to CalculateL2Time.
  
  karupayun
  authored and
  tensorflower-gardener
  committed
  f138930
- [XLA:GPU] Ignore `tileiras` older than CUDA 13.3 in the TensorIR emitter.
  
  ilya-biryukov
  authored and
  tensorflower-gardener
  committed
  9c4d5af
- [XLA:GPU] Propagate GPU errors to local collectives and coordination service.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  64a66d9
- Use _safe_print for all remaining console output in diagnose_dll_load
  MaddipatlaChetan24
  committed
  13beb4d
- Merge branch 'master' into patch-27
  MaddipatlaChetan24
  authored
  0a334a9
- Parameterize testDatasetAlreadyBatchedOffsets for Eager and Graph modes
  MaddipatlaChetan24
  committed
  9761201
- Fix deferred restoration of scalar DVariables
  MaddipatlaChetan24
  committed
  7a2a442
- Reverts 596eca059dd448a67178f25e31e7ab34222b1aaa
  tensorflower-gardener
  committed
  021a44d
- [XLA] Extract NearComparatorBase in xla/literal_comparison.cc to reduce template bloat
  
  cezheng
  authored and
  tensorflower-gardener
  committed
  c876640
- [XLA:GPU]: Support for tiling variadic reduce-scatter.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  900ad68
- [XLA:GPU] Only pass CUDA_GRAPH_INSTANTIATE_FLAG_USE_NODE_PRIORITY when graph nodes set non-default priority
  tensorflower-gardener
  committed
  2317429
- Fix IPv6 fallback regression and missing break statements
  MaddipatlaChetan24
  committed
  5ed35c4
- Add initialization check in AssignUpdateVariableOp and expand tests
  bodapatisaikrishna
  committed
  06c8af8
- Fix scalar tf.TensorShape([]) edge case in deferred DVariable restore
  MaddipatlaChetan24
  committed
  8ebbb6d
- Revert double-free regression in ScopedTFBuffer.__del__
  MaddipatlaChetan24
  committed
  9218c9a
- Integrate LLVM at llvm/llvm-project@bb80dc2a4f6f
  tensorflower-gardener
  committed
  f86884b
- [XLA:GPU] Migrate MultiGpuBarrier kernel cache and CollectiveMemoryCache to PerDeviceState.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  d079496
- [XLA:GPU] Move FlattenCollectiveFusion to collective_fusion.
  
  olegshyshkov
  authored and
  tensorflower-gardener
  committed
  39f393f
- Force GitHub PR sync
  MaddipatlaChetan24
  committed
  f2072e9
- Avoid driver-side waits on compute work in CC chunked staged transfers.
  
  vrishabmadduri
  authored and
  tensorflower-gardener
  committed
  fc058b6
- Add initialization checks in resource scatter and gather ops
  bodapatisaikrishna
  committed
  0e1fd5f
- Restore --xla_cpu_experimental_ynn_fusion_type= in saved_model_aot_compile.py to fix tfcompile ynn-fusion failure
  
  dmiltr3
  authored and
  tensorflower-gardener
  committed
  803243a
- Look through bitcasts when checking if multiply operands are worth fusing as inputs.
  
  vwbaker
  authored and
  tensorflower-gardener
  committed
  f825dfd
- Merge pull request #125541 from madib06ops:variant-reader-empty-metadata
  tensorflower-gardener
  committed
  c04db97
- Catch ValueError alongside InvalidArgumentError in roll OOB tests
  kaivalya-cyber
  committed
  2fe47e3
- PR #50663: [XLA:GPU][oneAPI] Align XPU Triton hook with upstream LLVM
  
  akhilgoe
  authored and
  tensorflower-gardener
  committed
  63149a5
- [XLA:GPU] Support moving f8e8m0fnu values around in Triton fusions.
  
  vwbaker
  authored and
  tensorflower-gardener
  committed
  b265374
- Merge pull request #128356 from GodlyDonuts:codex/fix-unigram-sampler-range-limit
  tensorflower-gardener
  committed
  efa7e6d
- Merge pull request #123855 from vishwakt:fix-concatenate-docstring
  tensorflower-gardener
  committed
  8aadd2f
- Set memory_kind in IFRT Proxy CopyArrays when it differs from source.
  
  lukebaumann
  authored and
  tensorflower-gardener
  committed
  67336f3
- [XLA:GPU] Fix GpuExecutableTest.InvokesDeviceErrorCallbackOnStreamError on ROCm.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  c0987b4
- [XLA] Size tuple results in the HLO isolator's default module size estimate
  tensorflower-gardener
  committed
  2fc67f7
- [XLA:GPU] Move collective launch dimension calculations to collective_emitter.
  
  olegshyshkov
  authored and
  tensorflower-gardener
  committed
  1c65ffa
- Add direct lowering support for linear, gather, and quantization ops.
  tensorflower-gardener
  committed
  27f95ae
- Subtract squared residual mean before reduction in tf.nn.moments
  
  dmiltr3
  authored and
  tensorflower-gardener
  committed
  91e6612
- Shard bias_op_test across 4 shards to prevent TSAN timeouts.
  
  dmiltr3
  authored and
  tensorflower-gardener
  committed
  4f8f953
- Add v4 GPU AOT goldens for the collective ops tests
  
  khasanovaa
  authored and
  tensorflower-gardener
  committed
  4db8337
- Add Python 3.14 PyPI classifier, Bzlmod toolchain, and 2.22.0 release notes.
  tensorflower-gardener
  committed
  b5d3f04
- [XLA] Copy instead of move async results in the evaluator
  tensorflower-gardener
  committed
  97d83d8
- [XLA:GPU] Fuse scaled dots with GemmFusionV2.
  
  vwbaker
  authored and
  tensorflower-gardener
  committed
  ba721bc
- [SDY][perf] Defer SymbolTable construction in SdyRoundTripDedupMeshesPass.
  
  Varcho
  authored and
  tensorflower-gardener
  committed
  3fa7838
Commits on Oct 15, 2026
- Integrate LLVM at llvm/llvm-project@2b35fb9283a4
  tensorflower-gardener
  committed
  c66b5ee
- Register Shardy pointwise sharding rules for elementwise MHLO ops.
  
  bixia1
  authored and
  tensorflower-gardener
  committed
  7d96cfa
- Ensure LaunchAsyncOpKernel and FallbackBatchResource invoke DoneCallback on all error paths.
  tensorflower-gardener
  committed
  3ce1681
- Harden lookup kernels and LiteRT model verification.
  
  snnn
  authored and
  tensorflower-gardener
  committed
  eba5052
- security/fuzzing: silent reporter, add.bin seed, stricter input zeroing
  cybrdude
  authored
  c0f8e26
- security/fuzzing: seed data and resource_loader dep for the TFLite fuzz target
  cybrdude
  authored
  ba730ad
- lite: initialize buffer_ptr before get_readonly_data
  cybrdude
  authored
  c1be76e
- Condense inline comment on scalar shard_info handling
  MaddipatlaChetan24
  committed
  4316853
- Revert uninitialized variable checks and prune test
  bodapatisaikrishna
  committed
  ad997db
- Fix float-to-double narrowing in TFLite quantized Mul kernel.
  
  veblush
  authored and
  tensorflower-gardener
  committed
  447c167
- Fix float-to-double narrowing in TFLite reference kernels.
  
  veblush
  authored and
  tensorflower-gardener
  committed
  5703b56
- Fix float-to-double narrowing in GetQuantizedConvolutionMultipler for int8/int16.
  
  veblush
  authored and
  tensorflower-gardener
  committed
  2a4f4d8
- [XLA] Fill elided HLO constants per parse, not per process. The HLO parser
  
  patrick-toulme
  authored and
  tensorflower-gardener
  committed
  b2cddb6
- Automated Code Change
  tensorflower-gardener
  committed
  f79e44d
- Harden convolution kernels and LiteRT model verification against int32_t overflows.
  
  snnn
  authored and
  tensorflower-gardener
  committed
  b8acc86
- Automated Code Change
  tensorflower-gardener
  committed
  8f3fbff
- Automated Code Change
  tensorflower-gardener
  committed
  3ae0bb8
- Merge pull request #128461 from kaivalya-cyber:fix/np-roll-axis-bounds
  tensorflower-gardener
  committed
  9ffcb9f
- [XLA] Don't include thread local allocations in total memory report
  
  vsytch
  authored and
  tensorflower-gardener
  committed
  16abeb8
- Clamp XLA MatrixDiagPart's actual lengths at 0 and widen its tests
  vishwakt
  committed
  9a62335
- Harden `SLICE` kernel and XNNPACK delegate validation and expand unit test coverage:
  
  snnn
  authored and
  tensorflower-gardener
  committed
  575d488
- Test rank 5 in testHighRank and name the last dimension
  vishwakt
  committed
  b5a93a5
- Merge pull request #128483 from genius-0963:fix-isclose-infinity-handling
  tensorflower-gardener
  committed
  e017294
- Merge pull request #128537 from Abhirup0:fix-ragged-row-splits-validation
  tensorflower-gardener
  committed
  0e0a5db
- Merge pull request #128970 from MaddipatlaChetan24:patch-66
  tensorflower-gardener
  committed
  1a365fb
- Merge pull request #128732 from MaddipatlaChetan24:patch-61
  tensorflower-gardener
  committed
  8301a9a
- Merge pull request #128727 from MaddipatlaChetan24:patch-57
  tensorflower-gardener
  committed
  80cf079
- Merge pull request #127712 from MaddipatlaChetan24:patch-27
  tensorflower-gardener
  committed
  5ec2d6c
- Address non-blocking review suggestions: ordering invariant test, defensive None check, collective_group_leader test
  MaddipatlaChetan24
  committed
  488d68d
- Merge pull request #129013 from alepacheco:harden-sequence-example-fill-pass
  tensorflower-gardener
  committed
  3e019d6
- Gather diagonals on GPU only for static XLA MatrixDiagPart inputs
  vishwakt
  committed
  7eefc14
- Use mock.patch.object instead of mock.patch for static-analysis friendliness
  MaddipatlaChetan24
  committed
  8a6d505
- Merge pull request #128705 from MaddipatlaChetan24:patch-56
  tensorflower-gardener
  committed
  6090c4c
- Check every batch when initializing a lookup table
  alepacheco
  committed
  1a1b41b
- Free the Go SessionOptions config copy when TF_SetConfig fails
  vishwakt
  committed
  648a70a
- Merge pull request #128968 from MaddipatlaChetan24:patch-64
  tensorflower-gardener
  committed
  5166fae
- Merge pull request #129063 from vishwakt:hide-curl-symbols
  tensorflower-gardener
  committed
  f14ec8c
- [tflite] Set hoist_constants_as_args=False in TFLiteJaxConverterV2
  
  gnecula
  authored and
  tensorflower-gardener
  committed
  971afd2
- Add XNNPACK unit test for variable assignment with no outputs.
  tensorflower-gardener
  committed
  fd4c4a7
- [XLA:GPU] Avoid grouping GPU reductions that only share broadcasted inputs.
  
  iumitakgun
  authored and
  tensorflower-gardener
  committed
  c5c6515
- Move zizmor dangerous-triggers ignore comment onto pull_request_target line.
  tensorflower-gardener
  committed
  9bd2dd0
- Merge pull request #128515 from MaddipatlaChetan24:patch-53
  tensorflower-gardener
  committed
  857d3a4
- Merge pull request #123123 from cybrdude:fuzz/tflite-interp-overflow-target
  tensorflower-gardener
  committed
  0cc39b9
- Merge pull request #128730 from MaddipatlaChetan24:patch-59
  tensorflower-gardener
  committed
  763bbb7
- Merge pull request #128149 from adi-IL:bot/tensorflow-127914
  tensorflower-gardener
  committed
  00669fc
- Automated Code Change
  tensorflower-gardener
  committed
  8adf67a
- Reverts ba721bca5b7ddbc26e5e4b5c25143ee3659e1774
  tensorflower-gardener
  committed
  4cda1bb
- Fix doubled-brace f-string bug in embedding_lookup_sparse error message
  MaddipatlaChetan24
  committed
  767dedf
- Match the shape error in the lookup table init tests
  alepacheco
  committed
  a0bb7c3
- Merge pull request #127955 from RohithPariki:fix/127842-tflite-calibrator-shared-input
  tensorflower-gardener
  committed
  39097fa
- [XLA:GPU] Partially rollback device_error_callback in GpuExecutable.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  fc2a52a
- PR #50746: [ROCm] Remove flaky attempts to not pollute rocm rbe cluster on test failures
  
  alekstheod
  authored and
  tensorflower-gardener
  committed
  4001488
- Depend on platform:refcount in the lookup table init test
  alepacheco
  committed
  9d3dc4d
- [XLA:GPU]: Add variadic xtile.reduce_scatter op and migrate reduce-scatter lowering.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  14c10e5
- PR #50775: Bump lit from 23.1.2 to 23.1.3
  
  dependabot[bot]
  authored and
  tensorflower-gardener
  committed
  ca4bce0
- [XLA:GPU] Restore scale swizzling and support checks for cuDNN kScaledDot fusions.
  
  loislo
  authored and
  tensorflower-gardener
  committed
  9f748cd
- PR #50733: [XLA] Fix scalar index iteration on 32-bit platforms
  
  anutosh491
  authored and
  tensorflower-gardener
  committed
  f683e94
- Integrate Triton up to b575c0c38d428a609e25487db64051a5fa7bfab8
  
  mydatascience
  authored and
  tensorflower-gardener
  committed
  a4dacdb
- PR #50776: Bump actions/upload-artifact from 4.4.3 to 7.0.2
  
  dependabot[bot]
  authored and
  tensorflower-gardener
  committed
  8928d3c
- PR #50677: [ROCm] Drop per-node hipFuncSetAttribute from command buffer kernel nodes
  
  magaonka-amd
  authored and
  tensorflower-gardener
  committed
  21df4b8
- PR #50772: Fix dynamic kernel counts on 32-bit targets
  
  anutosh491
  authored and
  tensorflower-gardener
  committed
  84a72f2
- - Replace the previous primary/fallback approach by testing a list of algorithm strategies and choosing the first one that fits the estimated memory limits (or the most memory-efficient one if all …
  tensorflower-gardener
  committed
  fb4a2e9
- Remove unnecessary LLVM CodeGen deps from `cpu_aot_loader`
  
  EusebioDM
  authored and
  tensorflower-gardener
  committed
  cc84f2e
- Remove --end-of-options from git checkout in bazel_diff
  
  beckerhe
  authored and
  tensorflower-gardener
  committed
  496630d
- Merge pull request #128729 from MaddipatlaChetan24:patch-58
  tensorflower-gardener
  committed
  3390cf7
- PR #50088: [XLA:GPU] ragged_all_to_all dev_kernel: fire remote GIN puts before LSA copy
  
  sfvaroglu
  authored and
  tensorflower-gardener
  committed
  075d0a4
- [XLA:GPU] Inline collective launch dimension calculations into Triton collective emitter.
  
  olegshyshkov
  authored and
  tensorflower-gardener
  committed
  7372c80
- Use a `size_t` instead of an `int` in `LoadLastBuildStep`.
  
  qukhan
  authored and
  tensorflower-gardener
  committed
  9367422
- [XLA] Fix while-loop buffer double counting in NoFragmentationStatsHeap.
  tensorflower-gardener
  committed
  5c8408c
- [XLA:CPU] Restore piecewise math.expm1 expansion when C++ bitcode intrinsics are unavailable.
  
  bchetioui
  authored and
  tensorflower-gardener
  committed
  1075230
- [XLA:GPU] Remove xla_gpu_unsupported_enable_triton_multi_output_fusion check from TilingSpace. Also disable corresponding tests.
  
  Moerafaat
  authored and
  tensorflower-gardener
  committed
  5e450b6
- [XLA:CPU] Improve float32 rsqrt precision to 1 ULP on CPU
  
  hawkinsp
  authored and
  tensorflower-gardener
  committed
  e9b22e2
- [CHLO] Avoid underflow in chlo.zeta Euler-Maclaurin correction term
  
  Jake VanderPlas
  authored and
  tensorflower-gardener
  committed
  eb03ef0
- [XLA] Use CloneWithNewOperands in HandleReverse for binary ops.
  
  Jake VanderPlas
  authored and
  tensorflower-gardener
  committed
  642635c
- Automated Code Change
  tensorflower-gardener
  committed
  42cb085
- Reverts 5c8408ccd09632ecf57bab8cd7dcac9c4dcdd12e
  tensorflower-gardener
  committed
  ba00b30
- [XLA:GPU] Support sequential dimension tiling in BlockLevelFusionConfig. This should allow representation for reduce and scan operations.
  
  Moerafaat
  authored and
  tensorflower-gardener
  committed
  ccc81fe
- Disable XLA auto-clustering in testResourceVariableDtypeMismatch
  bodapatisaikrishna
  committed
  a64db63
- Clear moved-from `flat_hash_map`s in `PullTable::Reset()` to fix ASAN crash in `streaming_test`.
  
  belitskiy
  authored and
  tensorflower-gardener
  committed
  11a4725
- [tflite] Constant-fold PAD with a constant input.
  tensorflower-gardener
  committed
  a4a8e1c
- [XLA] Do not simplify integer pow(A, -1) to 1 / A in AlgebraicSimplifier
  
  SamGinzburg
  authored and
  tensorflower-gardener
  committed
  e961b4d
- [CI] Add BCR mirror fallbacks and Windows JVM DNS caching to XLA CI.
  
  belitskiy
  authored and
  tensorflower-gardener
  committed
  ed7d056
- Integrate LLVM at llvm/llvm-project@1ee8805f01a6
  tensorflower-gardener
  committed
  7b7e9ae
- Fold f32, DRQ, and Weight-Only Conv2D, DepthwiseConv2D, TransposeConv, and FC ops.
  
  ablavatski
  authored and
  tensorflower-gardener
  committed
  d276856
- [XLA:GPU] Don't read the autotune cache when determinism is required.
  tensorflower-gardener
  committed
  50c5820
- Merge pull request #115111 from rudra00434:codex/tf-99162-ifexp-error
  tensorflower-gardener
  committed
  74fd456
- PR #50820: [XLA:GPU][oneAPI] Align XPU Triton hook with upstream Triton
  
  akhilgoe
  authored and
  tensorflower-gardener
  committed
  e7c099d
- Test NewSession with an unparseable SessionOptions.Config
  vishwakt
  committed
  09e4e18
- [XLA:GPU] Enable MXFP8 E5M2xE5M2 and MXFP4xMXFP4 in cuDNN block-scaled dot.
  
  loislo
  authored and
  tensorflower-gardener
  committed
  b6b866c
- Merge pull request #125651 from MinaIbrahim10:fix/py314t-data-grappler-safety
  tensorflower-gardener
  committed
  61c032a
- Add HloInstructionHasName matcher to hlo_matchers.h.
  
  sashabu
  authored and
  tensorflower-gardener
  committed
  72972b7
- [XLA] Trap subnormal flush to zero in HLO isolator
  
  amir-saadat
  authored and
  tensorflower-gardener
  committed
  18be746
- Expose PJRT C-API Megascale and multi-slice extension headers to Megascale users.
  
  guoren97
  authored and
  tensorflower-gardener
  committed
  f7e8dd9
- Refactor fatal logging to use _fatal function
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  9079ebe
- Fix formatting issues in heartbeat.py
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  7e9c5d1
- Update _fatal function to accept keyword arguments
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  98e0670
- Add unit tests for heartbeat module functionality
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  c97dfba
- Guarantee _fatal crashes even if logging.fatal raises; add BUILD rule
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  d9b8aff
- Improve control flow context management in TPU utility
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  77abf4c
- Address review: add regression test for control-flow context restore
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  8753640
- Mark decomposition subgraphs as delegation-skippable in YNNPackDelegate.
  
  gcarranza-1
  authored and
  tensorflower-gardener
  committed
  a05e6f7
- Move `core/tpu:tpu_compile_interface` and TPU embedding proto utils to XLA
  
  ddunl
  authored and
  tensorflower-gardener
  committed
  a14be9a
- Introduce DtypeDict and normalize dtype keys in PlatformInfo
  tensorflower-gardener
  committed
  745eaa3
- [XLA:CPU] Add vectorized sin and cos intrinsics for f32 and f64
  
  seantalts
  authored and
  tensorflower-gardener
  committed
  a665106
- Fold f32, DRQ, and Weight-Only Conv2D, DepthwiseConv2D, TransposeConv, and FC ops.
  tensorflower-gardener
  committed
  2c93301
- Fix complex GetDistanceErr and restore skipping extreme inputs in C128 Log test.
  
  ermilovmaxim
  authored and
  tensorflower-gardener
  committed
  20f0713
- Fix `NanoRtExecutable::ManagedTemp` inline buffer alignment under AddressSanitizer.
  
  belitskiy
  authored and
  tensorflower-gardener
  committed
  d36d9ee
- Weave 'multithreaded' through all tsl file system implementations.
  
  matthiaskramm
  authored and
  tensorflower-gardener
  committed
  21f3e33
- Merge pull request #126450 from adi-IL:fix/linspace-bfloat16-half
  tensorflower-gardener
  committed
  7200926
- Merge pull request #126520 from somtri:fix-94379-biasadd-nchw-rank3
  tensorflower-gardener
  committed
  f93ef55
- Prevent copy removal from eliding copies that overwrite live values.
  
  mkuperst
  authored and
  tensorflower-gardener
  committed
  58dd296
- [SDY][cleanup] Replace sdy maximal mesh terminology with single_device under `tensorflow/` .
  
  Varcho
  authored and
  tensorflower-gardener
  committed
  b5f3385
- Preserve `LD_LIBRARY_PATH` in `SubProcessTest.SetEnviron` for shared-sanitizer builds.
  
  belitskiy
  authored and
  tensorflower-gardener
  committed
  9f80dcb
- PickRematerializationCandidates previously called GetCostOfRecompute from scratch on the entire block vector each time a candidate block grew by one instruction. For block size B, this performed O(…
  tensorflower-gardener
  committed
  c5ecd9a
- Merge pull request #126252 from madib06ops:tensorlist-decode-validate-elements
  tensorflower-gardener
  committed
  4bd7b20
- Fix asan bug introduced in the test
  tensorflower-gardener
  committed
  899832f
- [SDY][perf][2/2] Disable folding and constant CSE in SdyRoundTripExportPipeline canonicalizer.
  
  Varcho
  authored and
  tensorflower-gardener
  committed
  54a5f77
- PR #3022: Add exp2 and log2 operations.
  
  hawkinsp
  authored and
  tensorflower-gardener
  committed
  09a8a9d
- [XLA Highway] Dynamic dispatch for transpose microkernels (not called yet)
  
  vmoros
  authored and
  tensorflower-gardener
  committed
  48d24df
- [XLA:GPU] Deprecate --xla_gpu_experimental_enable_tiling_propagation flag and proto field
  
  metaflow
  authored and
  tensorflower-gardener
  committed
  b69a193
Commits on Oct 16, 2026
- Merge pull request #128752 from 372046933:fix_broadcast_overflow
  tensorflower-gardener
  committed
  d725930
- Move `tpu_compilation_result.proto` to `xla/tpu`
  
  ddunl
  authored and
  tensorflower-gardener
  committed
  a9c5f6a
- Flatten tuple-shaped outputs when extracting HLO modules in hlo_decomposer.
  
  mkuperst
  authored and
  tensorflower-gardener
  committed
  06d718a
- Preserve HLO original values across while-loop concat code motion.
  
  jcai19
  authored and
  tensorflower-gardener
  committed
  4e83e5a
- Integrate StableHLO at openxla/stablehlo@e7876d86
  
  mrguenther
  authored and
  tensorflower-gardener
  committed
  42d6aae
- Increase shard_count from 8 to 16 for take_test to prevent TSAN timeouts
  
  dmiltr3
  authored and
  tensorflower-gardener
  committed
  362d6d8
- [Shardy] Match tuple and GTE indices in `getX64CombineOnFuncResultSharding`
  
  mkuperst
  authored and
  tensorflower-gardener
  committed
  3d492f1
- [XLA] Improve float32, float16, and bfloat16 precision of sinh on CPU and TPU.
  
  hawkinsp
  authored and
  tensorflower-gardener
  committed
  50c771f
- [XLA:CPU] Support `math.exp2` and `math.log2` in `vectorize_xtile`.
  
  hawkinsp
  authored and
  tensorflower-gardener
  committed
  2e95144
- Support non-sum unreduced axes across Shardy verification, reshard, and manual export
  
  yashk2810
  authored and
  tensorflower-gardener
  committed
  e3d5ca8
- Integrate OpenXLA Shardy at c29509bd0c1430beaa7e767c866a131492ab1139
  
  yashk2810
  authored and
  tensorflower-gardener
  committed
  046efd7
- Enforce 16-byte buffer alignment in FlatBuffer export.
  
  gcarranza-1
  authored and
  tensorflower-gardener
  committed
  2cb3f78
- Ensure natural alignment for read-only tensor buffers in `Subgraph::SetTensorParametersReadOnly`
  
  gcarranza-1
  authored and
  tensorflower-gardener
  committed
  0f316d5
- Refactor _split function to include stack_fn
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  d294b75
- Address review: add regression test for multi-axis sharded pack_tf_tensor
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  43b78a3
- Address review: move test-only imports to module level
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  8e4a2bf
- Fix polymorphic_function import and untraceable pack() test
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  351ffde
- Skip test_pack_tf_tensor_multi_axis_sharded on <4-device meshes
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  a6905d2
- Fix dead device-count assertion in TPU test skip
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  56261f0
- Fix uncalled device_type method in TPU device-count skip guard
  
  MaddipatlaChetan24
  authored and
  dmiltr3
  committed
  fca14dc
- [XLA] Require orthogonal replica groups in AllReduceReduceScatterReorder
  
  SamGinzburg
  authored and
  tensorflower-gardener
  committed
  387fbac
- Merge pull request #128731 from MaddipatlaChetan24:patch-60
  tensorflower-gardener
  committed
  a06496b
- ## Problem
  
  SamGinzburg
  authored and
  tensorflower-gardener
  committed
  7425690
- `setHandler` in XLA's `error_util`
  
  ddunl
  authored and
  tensorflower-gardener
  committed
  8309e67
- Merge pull request #128966 from MaddipatlaChetan24:patch-62
  tensorflower-gardener
  committed
  aa1d11c
- Merge pull request #125907 from alwaysprince05:fix-device-placement-validation-jit-compile
  tensorflower-gardener
  committed
  c09c98d
- Merge pull request #129138 from alepacheco:validate-every-lookup-init-batch
  tensorflower-gardener
  committed
  f5b3f9d
- Merge pull request #128971 from MaddipatlaChetan24:patch-67
  tensorflower-gardener
  committed
  61457cd
- Fix heap address leak in ResourceBase::MakeRefCountingHandleName
  y198nt
  committed
  43e3784
- Automated Code Change
  tensorflower-gardener
  committed
  3d89798
- [Mosaic] Enable 16-bit find_first_set tpu.all_reduce
  
  tlongeri
  authored and
  tensorflower-gardener
  committed
  b3815fd
- Automated Code Change
  tensorflower-gardener
  committed
  20c7528
- Bump zizmor from 1.31.0 to 1.31.1 in .github/workflows/zizmor.yml.
  tensorflower-gardener
  committed
  08aa694
- Merge pull request #129139 from vishwakt:fix-go-session-config-leak
  tensorflower-gardener
  committed
  f39e398
- Merge pull request #129136 from MaddipatlaChetan24:fix-ragged-embedding-fstring-bug
  tensorflower-gardener
  committed
  86159e0
- Fix heap address leak in ResourceBase::MakeRefCountingHandleName
  y198nt
  committed
  49826a8
- Support scalar operands in XTile kClamp and kSelect
  
  akuegel
  authored and
  tensorflower-gardener
  committed
  0736391
- Merge pull request #128516 from MaddipatlaChetan24:patch-54
  tensorflower-gardener
  committed
  b6a4079
- Keep dynamic dimension sizes in the XLA ReverseSequence kernel
  
  vishwakt
  authored and
  dmiltr3
  committed
  9538096
- Run the dynamic ReverseSequence test eagerly and fix it on GPU and TPU
  
  vishwakt
  authored and
  dmiltr3
  committed
  f0ead62
- Run the dynamic ReverseSequence test's eager check in eager mode
  
  vishwakt
  authored and
  dmiltr3
  committed
  c863dac
- [XLA] Move `HandleSelect` and `HandlePad` from `HloEvaluatorTypedVisitor` to `HloEvaluator`
  
  cezheng
  authored and
  tensorflower-gardener
  committed
  fe714b5
- Merge pull request #128465 from MaddipatlaChetan24:patch-50
  tensorflower-gardener
  committed
  b03f527
- PR #50350: [XLA:GPU] Compile cuDNN graphs in parallel-thread computations
  
  keshavb96
  authored and
  tensorflower-gardener
  committed
  7bc5c49
- Merge pull request #129059 from vishwakt:fix-xla-matrix-diag-part-dynamic
  tensorflower-gardener
  committed
  48f18d7
- Hold the GIL only briefly when consuming Python tracer events.
  
  adityakakarot
  authored and
  tensorflower-gardener
  committed
  3afba85
- [XLA:GPU] Simply support checks in collective kernel info.
  
  olegshyshkov
  authored and
  tensorflower-gardener
  committed
  032f1da
- Merge pull request #128982 from vishwakt:fix-xla-reverse-sequence-dynamic
  tensorflower-gardener
  committed
  1fdd6fd
- [XLA:GPU] SplitkRewriter: use dot operand sizes for K padding and follow shape ops at fusion edge.
  
  mooskagh
  authored and
  tensorflower-gardener
  committed
  c6e8dda
- [XLA:GPU] Add Triton collective emitter and lowering for AllToAll.
  
  PatriosTheGreat
  authored and
  tensorflower-gardener
  committed
  27f616e
- Merge pull request #129167 from y198nt:fix-resourcebase-heap-pointer-leak
  tensorflower-gardener
  committed
  91caaf7
- PR #50870: [XLA:GPU][oneAPI] Align XPU Triton hook with upstream LLVM
  
  akhilgoe
  authored and
  tensorflower-gardener
  committed
  99f8176
- [XLA:GPU] Triton: don't transpose TMA tiles if the layout only moves unit dims.
  
  mooskagh
  authored and
  tensorflower-gardener
  committed
  42403c3
- PR #50778: [ROCm] Add gfx1102, gfx1152 and gfx1153 to supported AMDGPU versions
  
  lucbruni-amd
  authored and
  tensorflower-gardener
  committed
  cb82efa
- Rollback of cl/999614913 and forward fix for x86/ARM64
  
  dmiltr3
  authored and
  tensorflower-gardener
  committed
  124aa5e
- Enable RBE for B200 benchmark builds
  
  beckerhe
  authored and
  tensorflower-gardener
  committed
  9a6d454
- [XLA] Relay control dependencies when merging sibling fusions
  
  liepieshov
  authored and
  tensorflower-gardener
  committed
  7fd513f
- [XLA:CPU] Deduplicate kernels of hoisted small while loops.
  
  seantalts
  authored and
  tensorflower-gardener
  committed
  9c72341
- [XLA] Support ShapeTracker in DotOperandDims and simplify bitcast tracking.
  
  mooskagh
  authored and
  tensorflower-gardener
  committed
  d6aa13b
- [XLA:GPU] Migrate GpuExecutableVaRemapAllocator to PerDeviceState.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  579f1ed
- Emit embedded constant buffers as LLVM arrays and strip debug info from embedded bitcode.
  
  cezheng
  authored and
  tensorflower-gardener
  committed
  1526bb9
- [XLA:GPU] Unify multi-process tests logic.
  
  PatriosTheGreat
  authored and
  tensorflower-gardener
  committed
  b7fe3ca
- [XLA:GPU] Replace TransposeFolding with GpuTransposeFolding.
  
  mooskagh
  authored and
  tensorflower-gardener
  committed
  4ab586c
- Automated Code Change
  tensorflower-gardener
  committed
  db2d6ef
- Migrate multi-process GPU all-gather tests to MultiProcessGpuTest.
  
  PatriosTheGreat
  authored and
  tensorflower-gardener
  committed
  5a9a312
- Merge pull request #128996 from vishwakt:fix-onednn-softmax-high-rank
  tensorflower-gardener
  committed
  cdff152
- Merge pull request #128701 from bodapatisaikrishna:fix/113145-resource-var-dtype-mismatch
  tensorflower-gardener
  committed
  da890cf
- SpmdBuilder::SetBroadcastDimsForReshape records which dimensions of a
  
  SamGinzburg
  authored and
  tensorflower-gardener
  committed
  eae7bc8
- [XLA:GPU]: Prepare collective fusion and runtime for variadic all-gather.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  5194b35
- Automated Code Change
  
  nglevin
  authored and
  tensorflower-gardener
  committed
  c4d1e6d
- PR #50881: [XLA:GPU] Add missing CUPTI V2 dropped-records stub
  
  zhenying-liu
  authored and
  tensorflower-gardener
  committed
  0ca2a04
- [XLA:SPMD][unreduced][cleanup] Remove obsolete ResolveReductionOpForSharding helper.
  
  Varcho
  authored and
  tensorflower-gardener
  committed
  3fa835e
- Wait for all pending transfers and execution futures on error in FunctionalHloRunner
  tensorflower-gardener
  committed
  96a5acc
- [XLA:GPU]: Refactor collective emitter helpers for variadic collectives.
  
  sohaibiftikhar
  authored and
  tensorflower-gardener
  committed
  aede00d
- In `spmd_partitioner.cc`, `HandleScatter` pads `indices` with `-1` before dispatching to scatter partitioning helpers only when `indices` is already unevenly sharded. When `indices` is initially `{…
  
  SamGinzburg
  authored and
  tensorflower-gardener
  committed
  e50b01d

Commits

Commits on Oct 14, 2026

Commits on Oct 15, 2026