[pull] master from tensorflow:master - #2374

Open

pull[bot] wants to merge 10000 commits into

xiaohanhuang:masterfrom

tensorflow:master

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

Commits on Sep 9, 2026

Suppress new Pyrefly findings before upgrading to v1.2.0

oprypin
authored and
tensorflower-gardener
committed
591f71a
Integrate LLVM at llvm/llvm-project@979b72237338
tensorflower-gardener
committed
4c73bd8
PR #47472: [oneDNN][TSL] Add AMX_FP8 CPU feature definition and CPUID detection

gaurides
authored and
tensorflower-gardener
committed
e2a845e
Automated Code Change
tensorflower-gardener
committed
0256b7d
[XLA:GPU] Defer CUDA VMM memory deallocations while stream capture is active.

hawkinsp
authored and
tensorflower-gardener
committed
444b7c8
Support compiling CPU HLO modules in xla_compile by configuring CPU target options in XlaCompileMain, supporting TargetMachineOptions in SymbolRepository via CpuBackendSpecificData, and outputting …
tensorflower-gardener
committed
7cdde53
Update workspace.bzl with patch.
tensorflower-gardener
committed
12df209
Materialize restored variables in ResourceManager in TFRT-IFRT serving.

xiaoxlu
authored and
tensorflower-gardener
committed
23451dc

Commits on Sep 10, 2026

Skip tests unsupported under XLA strict auto-JIT.

lrdxgm
authored and
tensorflower-gardener
committed
2cdf78b
Shard nn_test across 20 shards.

lrdxgm
authored and
tensorflower-gardener
committed
6336e4c
Handle NumPy 2.0 2D cross product changes and adjust tensordot tolerance in TF numpy tests.

lrdxgm
authored and
tensorflower-gardener
committed
a86383b
Fix Windows MSVC build breakage after updating the recent LLVM integration.

xinxinmo
authored and
tensorflower-gardener
committed
f4156ed
Replace depracated TF_ASSERT_OK_AND_ASSIGN, TF_EXPECT_OK, and TF_ASSERT_OK with ASSERT_OK_AND_ASSIGN, EXPECT_OK, and ASSERT_OK.
tensorflower-gardener
committed
bad1eea
Automated Code Change
tensorflower-gardener
committed
119dba8
Add support for FLOAT8_E8M0FNU data type in TFLite/LiteRT.

majiddadashi
authored and
tensorflower-gardener
committed
8dafbb9
Add support for FLOAT8_E8M0FNU data type in TFLite/LiteRT.
tensorflower-gardener
committed
8da8c17
Automated Code Change
tensorflower-gardener
committed
8bcde9f
Demote CPU-only GPU profiler tests by removing GPU and cuda-only tags from xla_cc_test

IllogicalMoose
authored and
tensorflower-gardener
committed
a0cad18
[mhlo] Disable convert folding for resource elements

majiddadashi
authored and
tensorflower-gardener
committed
eeb00a9
Automated Code Change
tensorflower-gardener
committed
2d15cfc
Merge pull request #126748 from AshiteshSingh:fix/softmax-vjp-translation-invariance
tensorflower-gardener
committed
12410d9
Merge pull request #124019 from qiu-tiandev:fix-issue-65317-round-complex-dtype
tensorflower-gardener
committed
fa9c40e
Merge pull request #124253 from qiu-tiandev:fix-issue-105297-thread-validation
tensorflower-gardener
committed
9d208b1
Merge pull request #126902 from cuishuang:master
tensorflower-gardener
committed
565ffbb
Automated Code Change
tensorflower-gardener
committed
100beab
Merge pull request #126866 from GodlyDonuts:codex/validate-eager-capture-dtype
tensorflower-gardener
committed
4d73c87
Automated Code Change
tensorflower-gardener
committed
9f11da7
Merge pull request #125178 from GodlyDonuts:codex/fix-sqrt-subnormal-gradient
tensorflower-gardener
committed
598538c
Merge pull request #126745 from VirajMishra1:fix-central-crop-logical-op
tensorflower-gardener
committed
312575a
Merge pull request #126204 from kaivalya-cyber:fix/np-validate-axis-bounds-2
tensorflower-gardener
committed
2873b7e
Merge pull request #126870 from Noxtimo:fix/diag-scalar-k-validation
tensorflower-gardener
committed
b5c49da
Merge pull request #111188 from tensorflow:dependabot/pip/wheel-0.46.2
tensorflower-gardener
committed
18409bb
Merge pull request #126895 from kaivalya-cyber:fix/np-isinf-complex-parity
tensorflower-gardener
committed
6cdcdeb
Merge pull request #126377 from BenjaminDEMAILLE:fix-pluggable-device-kernel-c-api
tensorflower-gardener
committed
6b2dbee
Merge pull request #126899 from Gosling-dude:fix/112139-dense-bool-onednn-segfault
tensorflower-gardener
committed
10dd366
Update XNNPACK in XLA

dsharletg
authored and
tensorflower-gardener
committed
5c0b7a5
Add pfor loop converters for `ResourceGather` and `ResourceGatherNd`.
tensorflower-gardener
committed
3174d84
Merge pull request #126824 from GodlyDonuts:codex/fix-silu-float64-second-derivative
tensorflower-gardener
committed
6f6fcf8
Merge pull request #126770 from hphng:profiler-crash-on-nullptr
tensorflower-gardener
committed
872ed72
Merge pull request #126871 from GodlyDonuts:codex/skip-zero-element-batch-copies
tensorflower-gardener
committed
a865c56
Merge pull request #126862 from somtri:fix-125544-slice-oob-size-constant-fold
tensorflower-gardener
committed
bde5fe3
Migrate TFLite GPU delegate test targets to requires-gpu-sm90

IllogicalMoose
authored and
tensorflower-gardener
committed
250ef45
Automated Code Change
tensorflower-gardener
committed
6e07150
PR #48623: [XLA:GPU][oneAPI] Align XPU Triton hook with upstream LLVM

akhilgoe
authored and
tensorflower-gardener
committed
e0f285c
[XLA:CI] Fix Windows clang-cl build failures in OpenMPDialect and MLIR SymbolTable.h.

xinxinmo
authored and
tensorflower-gardener
committed
1f0bc0c
Bazel changes to visiblity for internal targets

EusebioDM
authored and
tensorflower-gardener
committed
453cb87
[XLA:GPU] Emit reshape as reshape + broadcast in experimental fusion emitter

vickyliu-go4it
authored and
tensorflower-gardener
committed
180a0d0
Relax rhs non-contracting rules for transpose. I think this is similar to what was done in cl/932960738 & now works for V2 as well.

vwbaker
authored and
tensorflower-gardener
committed
041d89f
test(tf2xla): add regression test for EuclideanNorm under XLA
adi-IL
committed
b7f7e4e
fix(tf2xla): register EuclideanNorm reduction OpKernel
adi-IL
committed
2a03ee7
fix(tf2xla): add lib:math dep for EuclideanNorm MaybeConjugate
adi-IL
committed
032ab48
fix(jit): allowlist EuclideanNorm for XLALite
adi-IL
committed
c51805a
Merge pull request #127019 from MaddipatlaChetan24:patch-9
tensorflower-gardener
committed
cc1a8ed
Merge pull request #127020 from MaddipatlaChetan24:patch-10
tensorflower-gardener
committed
366d422
Merge pull request #125795 from MaddipatlaChetan24:patch-3
tensorflower-gardener
committed
7a590c6
Merge pull request #126805 from GodlyDonuts:codex/fix-softplus-second-derivative
tensorflower-gardener
committed
51a6bea
fix(math_ops): use custom gradient for float64 tanh to preserve precision without memory regression
AshiteshSingh
committed
cfd94e1
Merge pull request #126968 from MaddipatlaChetan24:patch-8
tensorflower-gardener
committed
b9ebc56
Merge pull request #127012 from BenjaminDEMAILLE:fix-windows-flatbuffer-export
tensorflower-gardener
committed
1ed6d4e
Merge pull request #126935 from adi-IL:fix-tf2xla-bessel-i0-i1
tensorflower-gardener
committed
957c552
Automated Code Change
tensorflower-gardener
committed
56abd54
Do not propagate call-level frontend attributes to returned caller operands in CallInliner.

Kingston Mandisodza
authored and
tensorflower-gardener
committed
326135b
Support static extraction of CUDA kernel attributes and single-architecture CUBINs

beckerhe
authored and
tensorflower-gardener
committed
f5fb956
[XLA] Add SM 10.7 (VR200) to known CUDA compute capabilities in CI build script.

PatriosTheGreat
authored and
tensorflower-gardener
committed
f6c7595
PR #47495: [ROCm] Migrate SMI device queries from rocm_smi to amd_smi

nurmukhametov
authored and
tensorflower-gardener
committed
f1b70b1
Move internal native custom-call handler allowlist to package_groups.bzl.

beckerhe
authored and
tensorflower-gardener
committed
cc9a90f
PR #48489: Force link statically specific targets even with dynamic_mode=fully

alekstheod
authored and
tensorflower-gardener
committed
4810bcb
PR #48445: [XLA] Handle variadic reduces in HostOffloader

kodlan
authored and
tensorflower-gardener
committed
9d74024
Iteratively reuse buffers during allocation hoisting.
tensorflower-gardener
committed
51c41c4
PR #48543: [ROCm] Fix silently dropped scales in Triton scaled-dot.

zoranjovanovic-ns
authored and
tensorflower-gardener
committed
8b7cc7c
[XLA] Use a bitset for input dependency tracking in TryRemoveDeadWhileParams

patrick-toulme
authored and
tensorflower-gardener
committed
5287903
Fix and enable support for unbounded dynamic shapes in XLA TPU compilation.

Ryo-not-rio
authored and
tensorflower-gardener
committed
87d6af2
Optimize HloRematerialization and TuplePointsToAnalysis compile time.
tensorflower-gardener
committed
759116b
Enable `xla_enable_hlo_sharding_v3` by default.

Kingston Mandisodza
authored and
tensorflower-gardener
committed
adf629d
pjrt_c_api_client: Remove "PJRT C API" from the platform version string.

pizzud
authored and
tensorflower-gardener
committed
f8b6c27
[XLA] Add benchmarks for online topological sort.

hawkinsp
authored and
tensorflower-gardener
committed
8b4502b
Fix OSS Bazel build breakage in stable HLO temporary.patch

xinxinmo
authored and
tensorflower-gardener
committed
223047b
Test empty ResourceScatterNd updates
kokol16
committed
940d813
Improve precision of float32 exp on CPU from 6 ULP to 1 ULP.

hawkinsp
authored and
tensorflower-gardener
committed
b3e160d
Remove inline comments from benchmarking workflow `uses` paths.

Andrew-Dame
authored and
tensorflower-gardener
committed
b3768ab
[CI] Use new LLVM19 toolchain on Windows.

belitskiy
authored and
tensorflower-gardener
committed
f594e16
Keep FindStart within the same computation for pipelined while loops in LatencyHidingScheduler.

amitsabne1
authored and
tensorflower-gardener
committed
072cfc7
Merge pull request #126936 from adi-IL:fix-tf2xla-euclidean-norm
tensorflower-gardener
committed
5d2d621
Internal change for toolchain.

dmiltr3
authored and
tensorflower-gardener
committed
7f273b4
Integrate LLVM at llvm/llvm-project@e30271b35d8d
tensorflower-gardener
committed
d982ef6
Suppress new Pyrefly findings before upgrading to v1.2.0

oprypin
authored and
tensorflower-gardener
committed
4bce908
Enable sin/cos/tan in YNNPACK in XLA

dsharletg
authored and
tensorflower-gardener
committed
d0149c8
[XLA:GPU] Replace colons with underscores in XlaHloRunner thread pool name.

loislo
authored and
tensorflower-gardener
committed
d4e0990
Support single-arch CUBIN and kernel attribute extraction in CUDA runtime kernel registry

beckerhe
authored and
tensorflower-gardener
committed
871172c
Simplify slice(DUS) to slice.

WillFroom
authored and
tensorflower-gardener
committed
9249619
Rollback of PR #125178

dmiltr3
authored and
tensorflower-gardener
committed
d42e094
Fix Windows build in slim_model_importer using portable std::ifstream

majiddadashi
authored and
tensorflower-gardener
committed
ddbc474
Preserve float64 log-softmax tail gradients
GodlyDonuts
committed
55c7df3
[XLA:CI] Revert Windows clang-cl workarounds after upstream LLVM MLIR fix.

xinxinmo
authored and
tensorflower-gardener
committed
5618387
Update TableGen definition for WriteTrainingPredictions op to include sample_rate attribute.
tensorflower-gardener
committed
a9ac72c
Eliminate redundant copy insertion for in-place dynamic-update-slice chains and async-start transitions in while loops with xla_disable_while_loop_copies.

amitsabne1
authored and
tensorflower-gardener
committed
260ba21

Commits on Sep 11, 2026

[XLA:GPU] Add --xla_gpu_enable_persistent_symmetric_memory flag for persistent symmetric memory copy elision.
tensorflower-gardener
committed
f45f6b7
Remove PjRtCpuDevice and PjRtCpuClient in favor of CommonPjRtClient alternatives to continue simplifying the backends.

pschuh
authored and
tensorflower-gardener
committed
6bcd370
Relocate Windows SDK and MSVC headers to host mount C:\sdk in Windows CI.

ratgr
authored and
tensorflower-gardener
committed
c148e64
[XLA] Replace online_topsort with non-intrusive implementation.

hawkinsp
authored and
tensorflower-gardener
committed
e0059b6
[XLA][HLO Value Tracking] Fix and update OriginalValue handling in ConditionalCodeMotion

jcai19
authored and
tensorflower-gardener
committed
418f88b
[XLA] Add missing direction comparison
tensorflower-gardener
committed
e098719
Consolidate open-source collective passes into xla/hlo/transforms/collectives/

toli-y
authored and
tensorflower-gardener
committed
ce5cca4
Add support for Adreno 8xx GPUs in TFLite GPU delegate.

futscdav
authored and
tensorflower-gardener
committed
766dd1a
Automated Code Change
tensorflower-gardener
committed
8dc23ed
Consolidate open-source collective passes into xla/hlo/transforms/collectives/
tensorflower-gardener
committed
57275f8
Slice SDPA mask and V against K-derived values instead of their own dummy inputs, so all kv extents share one symbol + refactoring to make change more localized.

vksnk
authored and
tensorflower-gardener
committed
bcf1014
Update XLA benchmarks to require JAX 0.11+ and Python 3.12+
tensorflower-gardener
committed
8705405
Merge pull request #126814 from GodlyDonuts:codex/fix-log-softmax-float64-gradient
tensorflower-gardener
committed
3b46b5e
Automated Code Change
tensorflower-gardener
committed
8e8bbec
Add float64 softmax gradient regression
GodlyDonuts
committed
6e3617a
Merge pull request #126972 from kokol16:fix/resource-scatter-nd-update
tensorflower-gardener
committed
8ea4ae9
Move --test_env flags from test to build in tensorflow.bazelrc

akuegel
authored and
tensorflower-gardener
committed
dd06d4b
Merge pull request #126747 from AshiteshSingh:fix/tanh-grad-float64-precision
tensorflower-gardener
committed
89dbed7
Merge pull request #125386 from SparshGarg999:fix-tf-stack-single-tensor-validation-125374
tensorflower-gardener
committed
c921ba7
Merge pull request #127081 from Rahul-s-007:fix-uniform-quantized-conv-output-feature-dimension-validation
tensorflower-gardener
committed
1233e4f
Merge pull request #127023 from MaddipatlaChetan24:patch-11
tensorflower-gardener
committed
86070cc
PR #47643: [XLA:GPU][oneAPI] Guard sycl free function in oneCCL

nhatleSummer22
authored and
tensorflower-gardener
committed
c99b570
Run XLA Linux x86 GPU CI on CPU runner pool and rename to T4

akuegel
authored and
tensorflower-gardener
committed
72a73aa
PR #48559: [ROCm] Enable GpuKernelTilingTest.ReductionInputTooLarge on theRock

hsharsha
authored and
tensorflower-gardener
committed
cec8f0a
PR #48498: [GPU] Fix cuDNN fusion layout

dlcompilers-infra-bot
authored and
tensorflower-gardener
committed
dd3737b
Support target compute capability in XLA FFI AOT custom calls

beckerhe
authored and
tensorflower-gardener
committed
cee0d07
Lower bitcast-convert layouts using post-layout tiling copies.

reichlfl
authored and
tensorflower-gardener
committed
4f770d4
Support constant folding for shift_left, shift_right_arithmetic, and
tensorflower-gardener
committed
8992f65
Optimize TryRemoveDeadWhileParams by replacing flat_hash_set with InlinedBitSet

Jacob Wahlgren
authored and
tensorflower-gardener
committed
d4218f3
Run JAX Linux x86 GPU CI on CPU runner pool and rename to T4

akuegel
authored and
tensorflower-gardener
committed
d6e6726
Reverts 8992f652848ec7d51e37d0476332b56c1e9980d2

stjerngren
authored and
tensorflower-gardener
committed
c10cecb
[XLA:CPU] Clarify comments and Cody-Waite constant names in GenerateVF32Exp.

hawkinsp
authored and
tensorflower-gardener
committed
c9ee8fb
Migrate ABSL_MUST_USE_RESULT to C++17 [[nodiscard]]
tensorflower-gardener
committed
97bf931
Internal BUILD change.

wuhao5
authored and
tensorflower-gardener
committed
044dd41
PR #9721: SLICE: validate static-output slice bounds without changing allocation type

endorphin13
authored and
tensorflower-gardener
committed
f70947f
Fix Free-Threading (no-GIL) concurrency data race on tf.Tensor._shape_val.
tensorflower-gardener
committed
a3d568a
Migrate tf2xla and tf_jit to use PjRtCompiler and PjRtClient

Peter Gavin
authored and
tensorflower-gardener
committed
92c7c52
[XLA:MSA] Skip tuples in MaybeCreateLoopValue.
tensorflower-gardener
committed
cfb5281
Disable collective_ops_aot_test on gb300.

IllogicalMoose
authored and
tensorflower-gardener
committed
f7c42bc
Resolve loop iteration placeholder #$ to iteration 0 in comparison graph.
tensorflower-gardener
committed
0857229
Migrate tf2xla and tf_jit to use PjRtCompiler and PjRtClient
tensorflower-gardener
committed
0f8856f
Remove methods and fields from PjRtStreamExecutorClient that no longer need to

pschuh
authored and
tensorflower-gardener
committed
4acac40

Commits on Sep 12, 2026

fix(math): correct xdivy gradient w.r.t. x at x=0
jdymitarai
committed
a93f32d

Commits on Sep 13, 2026

Update compat.py
MaddipatlaChetan24
authored
cecbb3c
Validate axis bounds in experimental.numpy cumsum and cumprod
kaivalya-cyber
committed
51d1343

Commits on Sep 14, 2026

Fix MatrixDiagV2 gradient for ignored diagonal padding
twelfthlabor
committed
19f7371
Fix int narrowing and overflow crash in the tf2xla Pad kernel
somtri
committed
0942233
Integrate LLVM at llvm/llvm-project@7024b9e1b423
tensorflower-gardener
committed
9366621
Squashing commits after sync tool breakage
tensorflower-gardener
committed
f1431f4
Merge pull request #124450 from Cyrax321:fix-doc-typos
tensorflower-gardener
committed
795f4a1
[IFRT IR] Add utility to get a pretty print of a program argument location

ICGog
authored and
tensorflower-gardener
committed
6c26634
[IFRT] Tag mlir programs in unit tests to get syntax highlighting

ICGog
authored and
tensorflower-gardener
committed
e95a3da
Add unit tests to show non-flat support for SpaceToBatchConverter pass.

ekayaaslan
authored and
tensorflower-gardener
committed
4e75da6
Skip "Individual definition events" tests when GOOGLE_CUDA is not defined.

pschuh
authored and
tensorflower-gardener
committed
495923f
Add support for FLOAT8_E8M0FNU data type in TFLite/LiteRT.

majiddadashi
authored and
tensorflower-gardener
committed
e1d58fb
Add PjRtHostMemoryForDeviceManager to the CommonPjRtClientImpl constructor.

pschuh
authored and
tensorflower-gardener
committed
2303e78
Add a control dep from token to the nullary op inside kCall in a new pass `NullaryFunctionWrapInliner`

yashk2810
authored and
tensorflower-gardener
committed
e2c3ea4
Handle nullary primitives in program_order. We do this by rewriting the nullary primitive as `jax.jit(lambda token: eqn.primitive.bind(**params))(token)` and then program_order adds the opt_barrier…

yashk2810
authored and
tensorflower-gardener
committed
217f57e
Add support for FLOAT8_E8M0FNU data type in TFLite/LiteRT.
tensorflower-gardener
committed
abf5a16

Commits on Sep 15, 2026

Ignore tiling when comparing layout for dynamic tensor at runtime.

mkhanna15
authored and
tensorflower-gardener
committed
ca58a88
Remove PjRtStreamExecutorDevice and StreamExecutorGpuDevice in favor of the

pschuh
authored and
tensorflower-gardener
committed
a8152db
Rollback of PR #126894

beckerhe
authored and
tensorflower-gardener
committed
11f6b67
[SDY][HloShardingV3][Shard-Map] Take the value's rank when converting an HLO sharding to Shardy. This unblocks representing shard map boundaries with native shardings instead of string frontend att…

Varcho
authored and
tensorflower-gardener
committed
5cf326c
Consolidate open-source collective passes into xla/hlo/transforms/collectives/

toli-y
authored and
tensorflower-gardener
committed
be3e12e
Merge pull request #127316 from twelfthlabor:fix/matrixdiagv2-grad-align-127235
tensorflower-gardener
committed
8aa1ebd
Merge pull request #127297 from kaivalya-cyber:fix/np-cumsum-cumprod-axis-validation
tensorflower-gardener
committed
62fa450
Merge pull request #127294 from jdymitarai:fix-xdivy-grad-zero-x
tensorflower-gardener
committed
e17bf46
Merge pull request #127295 from MaddipatlaChetan24:patch-12
tensorflower-gardener
committed
099945e
Add option to disable PRED bit-packing during Literal serialization

akuegel
authored and
tensorflower-gardener
committed
b791b29
Merge pull request #127324 from somtri:fix-116716-tf2xla-pad-overflow
tensorflower-gardener
committed
412b473
[SDY] Enable full version of explicit reshards by default.

Varcho
authored and
tensorflower-gardener
committed
0fb31ed
[XLA:CPU] Read scatter operand from loop-carried output tensor

akuegel
authored and
tensorflower-gardener
committed
a757345
[SDY] Enable full version of explicit reshards by default.
tensorflower-gardener
committed
b922306
Add unit tests for non-flat support on BroadcastCanonicalizer.

ekayaaslan
authored and
tensorflower-gardener
committed
29286b7
[XLA:CPU] Read scatter operand from loop-carried output tensor
tensorflower-gardener
committed
2d4e9b7
Fix left-alignment in MatrixDiagPartV2 and MatrixSetDiagV2 gradients
twelfthlabor
committed
e0cf3cf
Migrate the deprecated ABSL_DEPRECATED to C++14 [[deprecated]]
tensorflower-gardener
committed
72fa946
Automated Code Change
tensorflower-gardener
committed
d90d5b1
Automated Code Change
tensorflower-gardener
committed
011273a
Automated Code Change
tensorflower-gardener
committed
bdc7610
Migrate the deprecated ABSL_DEPRECATED to C++14 [[deprecated]]
tensorflower-gardener
committed
1857695
Automated Code Change
tensorflower-gardener
committed
acff531
Automated Code Change
tensorflower-gardener
committed
a37deb5
Merge pull request #108937 from AbhishekChaudharii:issue_52195
tensorflower-gardener
committed
8b86575
[XLA] Store LayoutAssignment buffer constraints densely by LogicalBuffer id

patrick-toulme
authored and
tensorflower-gardener
committed
38574fc
Merge pull request #126063 from VaggelisGian:fix-elu-second-derivative
tensorflower-gardener
committed
db48d27
Update RELEASE.md on master for v2.23.0
tejaswin432
committed
5880e12
Merge pull request #126823 from GodlyDonuts:codex/fix-softmax-float64-gradient
tensorflower-gardener
committed
0f2eede
Update version numbers to 2.22.0-rc0
tejaswin432
committed
7926358
Add a doc for large tflite models.

snnn
authored and
tensorflower-gardener
committed
31d94c8
[SDY][HloShardingV3][Remove-Opt-Out] Support HloShardingV3 for shard maps without frontend attributes.

Varcho
authored and
tensorflower-gardener
committed
5686172
[XLA] Make DfsHloVisitor::ResetVisitStates O(1) with a generation counter

patrick-toulme
authored and
tensorflower-gardener
committed
3f18eae
Propagate context in Future::Map and MakeFutureOn

junwhanahn
authored and
tensorflower-gardener
committed
6ec23fa
[XLA] Precompute HloValue uses in one linear pass per dataflow analysis

patrick-toulme
authored and
tensorflower-gardener
committed
65b6ec6
[XLA:GPU] Add a flag to gate using RAFT for stable TopK.

apivovarov
authored and
tensorflower-gardener
committed
51e7b78
[Mosaic] tpu.memref_reshape: generalize support for contiguous tiled layouts without nested tiles

tlongeri
authored and
tensorflower-gardener
committed
8bf6428

Commits on Sep 16, 2026

Add using stmt for overloaded methods which shadow overloads in PjRtClient with

pschuh
authored and
tensorflower-gardener
committed
4d2af11
[XLA] Preserve aliasing on kAsyncUpdate during proto roundtrip

vsytch
authored and
tensorflower-gardener
committed
c678f17
[SDY][Cleanup][Bug-Fix] Clean up verbose JAX-user error messages for Shardy.

Varcho
authored and
tensorflower-gardener
committed
b5bd1de
StreamExecutor: Skip PCIe bandwidth check for GB300 and remove fake PCIe fallback.

IllogicalMoose
authored and
tensorflower-gardener
committed
473df75
Replace assert with ValueError for axis validation in np.compress
kaivalya-cyber
committed
4a24079
Remove unused gpu_event_mgr.cc.

Peter Gavin
authored and
tensorflower-gardener
committed
a66226f
Unify tf_cuda_tests_tags, tf_cuda_base_tests_tags, and tf_cuda_2gpu_tests_tags in build_config_root.bzl

IllogicalMoose
authored and
tensorflower-gardener
committed
2b7f73c
Prefactor: Rename NeedsMirroredAllocation to ShouldBeMirrored in MsaAlgorithm and clarify with necessary source code comments.

rishipal
authored and
tensorflower-gardener
committed
7754cfa
Merge pull request #127445 from tensorflow:update-release-md-2.23.0
tensorflower-gardener
committed
9b3a60e
Merge pull request #127444 from tensorflow:version-numbers-2.22.0rc0
tensorflower-gardener
committed
5018700
StreamExecutor: Fix C++17 compatibility in cuda_executor_test.

bhatuzdaname
authored and
tensorflower-gardener
committed
155ff28
Merge pull request #126685 from ankitxvx:fix-image-projective-transform-xla
tensorflower-gardener
committed
36008ee
Expose and enable dataflow-based input generation by default in HLO isolator.

bhatuzdaname
authored and
tensorflower-gardener
committed
c25741f
Decouple cuda_tile and tensor_ir patch lists into series.bzl to allow autonomous patch updates.

chsigg
authored and
tensorflower-gardener
committed
bca3909
Automated Code Change
tensorflower-gardener
committed
ef923f4
Set unreduced result sharding when inserting sdy.all_reduce for reduction factors in insert-explicit-reshards.

bixia1
authored and
tensorflower-gardener
committed
a007ef3
Merge pull request #127408 from twelfthlabor:fix/matrix-diag-part-v2-grad-sibling
tensorflower-gardener
committed
fbde951
PR #47813: [XLA:GPU][oneAPI] Add utility file for oneDNN convolution implementation

kanvi-nervana
authored and
tensorflower-gardener
committed
f33f420
Replace dynamic device memory allocations in RAFT select_k execution with a

apivovarov
authored and
tensorflower-gardener
committed
566239b
[XLA:CPU:Autotuner] Add `BlockLevelEmitter` autotuner backend.

penpornk
authored and
tensorflower-gardener
committed
03607cc
[XLA:GPU] Don't claim CompileAndLink support without nvlink
tensorflower-gardener
committed
bb5d2fc
Add symmetric bound splitting for Add/Subtract in ConstraintPropagator

bhatuzdaname
authored and
tensorflower-gardener
committed
58fa194
PR #48738: [XLA:GPU] Clear sharding of fusion instructions in hoist-fused-bitcasts

dlcompilers-infra-bot
authored and
tensorflower-gardener
committed
f34cc64
[XLA:CPU] Add benchmarks for transpose for f16/bf16/f64.

pifon2a
authored and
tensorflower-gardener
committed
acef17a
PR #48404: Fix slice of reshape to reshape of slice when reshape operand has leading unit dimensions

Pangoraw
authored and
tensorflower-gardener
committed
2865c89
PR #48369: [ROCm] Lower bf16 exp/log via f32 transcendentals on all AMD GPUs

zoranjovanovic-ns
authored and
tensorflower-gardener
committed
6ce3e0e
PR #48747: [XLA:GPU][codegen] Use driver intrinsics to lower math.cbrt in the SPIR-V backend

akhilgoe
authored and
tensorflower-gardener
committed
7a1b341
PR #47989: [oneDNN] Add HLO for keras vision transformer model with jax backend for BF16

gaurides
authored and
tensorflower-gardener
committed
43c2b42
[XLA:CPU] Add expansion patterns for 8x16 vector transposes.

pifon2a
authored and
tensorflower-gardener
committed
d5cf9a5
PR #48776: [ROCm] Switch build spawn strategy to dynamic mode for rbe

alekstheod
authored and
tensorflower-gardener
committed
9d48b4c
Reverts bca39096b90109e9c73f44f7c8a6da8f46b24091

beckerhe
authored and
tensorflower-gardener
committed
a545c74
PR #48879: Fix iteration in the reduce window rewriter.

dlcompilers-infra-bot
authored and
tensorflower-gardener
committed
d4c5c7e
PR #48906: [ROCm] Update reference to the latest rules_ml_toolchain

alekstheod
authored and
tensorflower-gardener
committed
281cb6b
[XLA:GPU] Fix scan carry indexing for multi-dimensional scans.

chsigg
authored and
tensorflower-gardener
committed
4c67bc4
PR #48815: [ROCm] Fix hipcc path for Bzlmod

afzpatel
authored and
tensorflower-gardener
committed
1cf4e90
Automated Code Change
tensorflower-gardener
committed
f7a35c1
Automated Code Change
tensorflower-gardener
committed
19e7cc8
Automated Code Change
tensorflower-gardener
committed
1eca80e
PR #48810: Fix DWYU presubmit target selection and TSL repository resolution

ezhulenev
authored and
tensorflower-gardener
committed
102a566
[XLA] Add a CI check prohibiting std::string_view in favor of absl::string_view.

thcmbs
authored and
tensorflower-gardener
committed
8885e54
Update LLVM SHA-256 to match mirror (#127498)
beckerhe
authored
807664d
Dont propagate through calls for dataflow analysis on ScanLoopAccumulatorInputUnification.

ekayaaslan
authored and
tensorflower-gardener
committed
74ba125
Add clarifying comments for HloEvaluator around caching and child evaluators.

ekayaaslan
authored and
tensorflower-gardener
committed
7f12579
Merge pull request #127296 from kaivalya-cyber:fix/np-compress-axis-validation
tensorflower-gardener
committed
1467e71
[XLA:CPU] Set the contract fast-math flag on emitted floating point arithmetic.

seantalts
authored and
tensorflower-gardener
committed
34d7ab2
[SDY][HloShardingV3][Remove-Opt-Out] Support HloShardingV3 for shard maps without frontend attributes.
tensorflower-gardener
committed
f5f88cf
Add copy-to-host support for arrays with non-addressable devices.
tensorflower-gardener
committed
4b8fd27
Link JAX CPU kernels into xla_compile, enabling CPU custom call and typed FFI handler registrations (such as LAPACK, sparse, and tridiagonal solve) during compilation.
tensorflower-gardener
committed
be02756
[XLA Highway] Expand TransposeMicroKernel unit test coverage
tensorflower-gardener
committed
1362809
[XLA:CPU] Lower threshold for minor-dimension concatenate fusion to enable memcpy

seantalts
authored and
tensorflower-gardener
committed
3cca0ea
Enable migration from the deprecated ABSL_ATTRIBUTE_UNUSED attribute to C++17 [[maybe_unused]]
tensorflower-gardener
committed
f1fcf93
Add general event-tracing API to PjRtRawClient.

pschuh
authored and
tensorflower-gardener
committed
988038e
Revert kSpmdBoundaryCopyAttr copy removal in SpmdPartitioner.

bixia1
authored and
tensorflower-gardener
committed
37eca37
suppress pyrefly errors related to JAX scalar types

Jake VanderPlas
authored and
tensorflower-gardener
committed
e511b24
[XLA:GPU] Count leaf thunks against xla_gpu_graph_min_graph_size in CommandBufferConversionPass

hawkinsp
authored and
tensorflower-gardener
committed
8b9b213
[Mosaic] Relax tpu.mask_cast verifier to allow SC shapes, and emit error message instead of silent failure

tlongeri
authored and
tensorflower-gardener
committed
69161e2
Stop the HLO isolator reporting false-positive numerical mismatches
tensorflower-gardener
committed
aa4b55b
[XLA:GPU][Autotuner] Add support for a preferred autotuner backend in XLA GPU.

derdrdirk
authored and
tensorflower-gardener
committed
b83b38d