hash
stringlengths
40
40
msg
stringlengths
1
131k
author
stringlengths
1
33
email
stringlengths
0
57
date
int64
1,447B
1,698B
a3ebc06f46d940d2bf37b370d7eb29335f4b6c48
compat: Update forward compatibility horizon to 2023-09-05 PiperOrigin-RevId: 562705621
A. Unique TensorFlower
gardener@tensorflow.org
1,693,904,533,000
a442440b75f3a87083281dc1fd49589a9020c72e
Fix the build by adding a missing include PiperOrigin-RevId: 562718621
Dmitri Gribenko
dmitrig@google.com
1,693,908,511,000
2f2123b4e8bd00d4a13e672abc2cc805c13cc248
[lite] Update sample_stable_delegate for promotion of acceleration/configuration from experimental. In TensorFlow 2.13, lite/experimental/acceleration/configuration was moved to lite/acceleration/configuration, and although the old location is currently still supported for backwards compatibility, sample code should b...
Fergus Henderson
fergus@google.com
1,693,921,156,000
f564fa920c4ef55090128130897bdf1a6f682ba8
[XLA:GPU] Simplify operand bytes read computation in ProducerInputAccessTime. * Simplify expression. * Add comments. * Avoid special case for 0/0 divisions situation. PiperOrigin-RevId: 562762716
Oleg Shyshkov
shyshkov@google.com
1,693,921,313,000
78adbabe0da7b8d657c3b43fe836d14f28330ce4
Clean up aliased targets and point to real build targets PiperOrigin-RevId: 562762976
Mason Chang
masonchang@google.com
1,693,921,392,000
e342783318459a6e9ea37ea01f3b0fc3ba6743e4
Add `main()` calls to some tests, fix up broken ones. These tests weren't being run before, due to the `main` method not being called. PiperOrigin-RevId: 562763593
A. Unique TensorFlower
gardener@tensorflow.org
1,693,921,604,000
dba94edd570941dfa4d58c41565268b70e86b390
Integrate LLVM at llvm/llvm-project@12a789710e2b Updates LLVM usage to match [12a789710e2b](https://github.com/llvm/llvm-project/commit/12a789710e2b) PiperOrigin-RevId: 562770611
Dmitri Gribenko
dmitrig@google.com
1,693,923,710,000
ff7923b899f632f7efad8a5ea501b8596feed0cd
Add an overloaded version of `CreateBuffersForAsyncHostToDevice` that takes `PjRtMemorySpace` in `xla::PjRtClient`. PiperOrigin-RevId: 562774929
A. Unique TensorFlower
gardener@tensorflow.org
1,693,924,824,000
644b1cd908326a1b66b0c79f3a9d03ac59b67bea
[XLA:GPU] Disable GPU runtime for autotuning compilation Excluding the reference executables, because they can be more complex, and the runtime based pipeline is supposed to be more stable. This saves host RAM during compilation. PiperOrigin-RevId: 562782335
Tamás Danyluk
tdanyluk@google.com
1,693,926,649,000
735f5f200274ffce220f6b1047e267b316040abf
[XLA:GPU] Make storage of new fragments safer Pointers to elements of std::vector can be invalidated after each push_back/emplace_back if it triggers a reallocation. std::list doesn't have this problem. PiperOrigin-RevId: 562783331
Tamás Danyluk
tdanyluk@google.com
1,693,926,897,000
5d887db2b875fb0369d645b6d466bc3853c536dd
Remove 64bit->32bit cast in step id generation as host callback now supports 64bit. PiperOrigin-RevId: 562792441
Kuangyuan Chen
chky@google.com
1,693,929,111,000
4ea07593c071931b31f380c91b81e40dd10efcc0
[XLA:GPU] Speed up running light outside compilation ops. Creating an OpKernel on each call is very slow. So we add a hack to cache these. In fact, we leak them permanently! The idea is to put a 128-bit hash of the TfCallbackData into the `opaque` of the custom-call. We use this to look up the OpKernel, in a hasht...
Justin Lebar
jlebar@google.com
1,693,934,039,000
038720154aa730362c8812cc73dbb02414f81d98
Revert changes made to the RemoveReshapeBeforeFullyConnected pattern. PiperOrigin-RevId: 562818619
A. Unique TensorFlower
gardener@tensorflow.org
1,693,934,451,000
e0dae969378f3043d687d5f07d1c054d9ccabd3f
Drop legacy keras usages PiperOrigin-RevId: 562819796
Faizan Muhammad
fmuham@google.com
1,693,934,669,000
8ebd0f82589f39bc01e4598696349d9f7f0a50ff
Augments the problem formulation to limit the # of "departures" from the trivial sharding strategy. PiperOrigin-RevId: 562823026
A. Unique TensorFlower
gardener@tensorflow.org
1,693,935,251,000
8da51494ffc4f5355dba143e6f460d1433b5f45b
Invert the "is inlined" bit of absl::Status This change makes RepToPointer/PointerToRep have 0 instructions. This makes IsMovedFrom simpler (although this could always have left out the IsInlined check since that bit can never be set on the aligned pointer) In exchange, it makes CodeToInlinedRep slower, but does not...
A. Unique TensorFlower
gardener@tensorflow.org
1,693,935,953,000
de9386a491d9821d7f722edd77acd44f1067e7db
Add decorators for running a test or test class in V1 and V2 modes. PiperOrigin-RevId: 562856608
A. Unique TensorFlower
gardener@tensorflow.org
1,693,941,968,000
b4696c97e70682bf698dd7217323f140de1d5381
Update OSS ml_dtypes library. Required for int4. PiperOrigin-RevId: 562857660
Antonio Sanchez
cantonios@google.com
1,693,942,219,000
528ad01d26ae47ef9452bf2ef6fd49e9c0f4f2ba
Restrict phase 1 default visibility to package default of MLIR friends instead of public PiperOrigin-RevId: 562870817
Mason Chang
masonchang@google.com
1,693,945,065,000
a68c7be9ca91117b3c0b10a33f2da0eb91ad07ca
Temporarily disable xnn_enable_arm_i8mm before clang toolchain upgrade. PiperOrigin-RevId: 562884598
Weiyi Wang
weiyiw@google.com
1,693,947,820,000
dd1d02228720957171d5da79e6de7da9b693e9fd
Add a streamz to record how often a MLIR Module has a TPUPartitionedCallOp. This will be used and inserted into the V1CompatBridge to verify that TPUPartitionedCallOp == Inference Graph. PiperOrigin-RevId: 562889821
Mason Chang
masonchang@google.com
1,693,948,811,000
4acc962d688500b656ae1b696ced2aeaf740823d
Enable new weight-only pipeline The default pipeline for weight-only quantization is migrated to the newly added passes. This change temporarily disable per-channel quantization tests as it will be added in the follow-up CL for the new passes. PiperOrigin-RevId: 562897980
Songyi Han
syhan@google.com
1,693,950,411,000
c3a9428e1c9be6c8133920f4e74dc38474247ca7
Add inline annotation providing auxiliary information for source code indexing. The annotation indicates which pybind module calls a certain function. PiperOrigin-RevId: 562906575
Daniel Chen
chencm@google.com
1,693,952,148,000
aea2f3954a5437635fb9239853edd3929d605d8e
Merge pull request #60749 from mgokulkrish:unittest_fix PiperOrigin-RevId: 562913768
TensorFlower Gardener
gardener@tensorflow.org
1,693,954,041,000
eda7db090d8699a589dc8190a07c6ebabab683dc
Make couple of TFRT library targets visible to public. PiperOrigin-RevId: 562919675
A. Unique TensorFlower
gardener@tensorflow.org
1,693,955,087,000
f05ae8263f2675b539c88b8f03bbb178eb152dcc
[TFLite] Fix cmake build kernel test fail Using CMake to build the TFLite kernel test will fail in linking stage when `-DTFLITE_ENABLE_GPU=OFF` ld: libtensorflow-lite-test-base.a(op_version.cc.o): in function `.LBB1_18': op_version.cc:(.text._ZN6tflite15UpdateOpVersionEPh+0x14c): undefined reference to `tflite::GetOp...
Bruce Lai
bruce.lai@sifive.com
1,691,137,987,000
112769b8cc750ce729f86fc914c83fdbd0fe4931
update sqlite from 3.40.1 to 3.43.0
justkw
karen.wu@intel.com
1,693,970,998,000
892d28b4c9cb20f61313dae2412c11c2412ef514
Fix bug introduced by cl/562816602 in light outside compilation. If we run the same op twice with two different output buffers, it would reuse the old output buffers! Oops. :) PiperOrigin-RevId: 562978988
Justin Lebar
jlebar@google.com
1,693,974,129,000
e4c5b75765bd6ca7cca48b4e7e8a84ab04569f9f
[HloValueSemanticsAnalysis] Adds heuristics to correctly identify weight and activation gradients. PiperOrigin-RevId: 562988250
Jinliang Wei
jlwei@google.com
1,693,977,161,000
f303dcad8fbdbaa37c58bb6f7b963227ffbf3429
Merge pull request #61762 from elfringham:flaky_snapshot PiperOrigin-RevId: 562988717
TensorFlower Gardener
gardener@tensorflow.org
1,693,977,643,000
0017627325a48530c9a9abfe226d4943fe9fc542
Internal Code Change PiperOrigin-RevId: 562999284
A. Unique TensorFlower
gardener@tensorflow.org
1,693,980,484,000
dd05f19239d804ad6fcfb31431b1678bc1dc3b4f
Update XNNPack dependency - Bring in polyfill for _mm256_zextps128_ps256 intrinsic for old compilers PiperOrigin-RevId: 563017375
Marat Dukhan
maratek@google.com
1,693,985,949,000
2c860a38b03047a9750ab72a476df4e58d4c5ddb
Update GraphDef version to 1611. PiperOrigin-RevId: 563034213
A. Unique TensorFlower
gardener@tensorflow.org
1,693,990,942,000
afbc88e6197cf0051ec93c38725932173c2173c6
compat: Update forward compatibility horizon to 2023-09-06 PiperOrigin-RevId: 563034227
A. Unique TensorFlower
gardener@tensorflow.org
1,693,990,946,000
75200ba6f825e6d4a32e06306d9aff62308701a6
Promote the TF Lite C++ signature runner APIs from experimental to stable. As part of this, move the definition of tflite::Interpreter::GetSignatureRunner from interpreter_experimental.cc to interpreter.cc. (The other newly promoted experimental methods of Interpreter were already defined inline in interpreter.h.) P...
Fergus Henderson
fergus@google.com
1,694,001,918,000
c51e75db10baabcc56169842705302873a86ae52
[HloValueSemanticsAnalysis] Fix memory sanitizer failure introduced by cl/562988250. PiperOrigin-RevId: 563090539
Jinliang Wei
jlwei@google.com
1,694,008,249,000
2880a4bfe8fcc3023dac8e6192e0d020c84c9017
SHLO PTQ Migration: Add an opset for integrating into TF Quantizer as an option. StableHLO Quantizer will be available as an option in the TF Quantizer in StableHLO Quantizer v1. PiperOrigin-RevId: 563091486
Jiyoun (Jen) Ha
jiyounha@google.com
1,694,008,551,000
b1ca9aeca85ab1be6fcdc68624fce17131b1f3a0
Set default values in `PassConfig` at declaration point. PiperOrigin-RevId: 563097141
Quentin Khan
qkhan@google.com
1,694,010,182,000
40e1c15ed6c7eaf0c8ee20418af12840821e69f0
Merge pull request #61679 from elfringham:fix_ub PiperOrigin-RevId: 563099817
TensorFlower Gardener
gardener@tensorflow.org
1,694,011,227,000
e23c4445adbc98011c28b5dad89f07614273ca98
[XLA:GPU][NFC] Make FindHeroReduction public This util function will be needed to filter out reduction fusions PiperOrigin-RevId: 563109222
Son Tuan Vu
vuson@google.com
1,694,013,273,000
9e51a0d179f5b08ab6a8b7627b66b61396c52432
Remove no longer used StopChipHeartbeats functions. PiperOrigin-RevId: 563135458
A. Unique TensorFlower
gardener@tensorflow.org
1,694,019,383,000
6c999686769dc21cf6af1adf91ad7abfa2eadeae
[PJRT C API] Set the visibility of GPU plugin GetPjrtApi to be public. PiperOrigin-RevId: 563145338
Jieying Luo
jieying@google.com
1,694,021,490,000
cb247d6878b7c74c137b2296abe13481f85e819e
Move legalize_tf_quant_test to MOT folder The test exercises quantization ops. PiperOrigin-RevId: 563149441
A. Unique TensorFlower
gardener@tensorflow.org
1,694,022,317,000
318305231d2b1aa9a1bc3c4db54941b3108de978
[HloValueSemanticsAnalysis] Adds handler for Domain and handler for Sharding custom call. PiperOrigin-RevId: 563152879
Jinliang Wei
jlwei@google.com
1,694,022,893,000
bfe403cdf6d1c80c0f02ae4ba710df8291e3553a
Creating a C++ Tensor object on a python buffer will eat a reference to the buffer, so we need to increase their reference count. PiperOrigin-RevId: 563154924
Yifan Jiang
yifanjiang@google.com
1,694,023,219,000
abcc910915d3e561d8742a2d21349eea07e9c85f
Modify comparators to satisfy strict weak ordering in sorting PiperOrigin-RevId: 563158112
A. Unique TensorFlower
gardener@tensorflow.org
1,694,023,676,000
3eca8eda0fda368a1f8c401a7bdd2c45a9b0d8a2
Integrate LLVM at llvm/llvm-project@c0703eaec1df Updates LLVM usage to match [c0703eaec1df](https://github.com/llvm/llvm-project/commit/c0703eaec1df) PiperOrigin-RevId: 563167024
Dmitri Gribenko
dmitrig@google.com
1,694,025,333,000
7ed87b3e9dac52c84ababc1a4719e9543bd08c1f
Internal change only PiperOrigin-RevId: 563167485
A. Unique TensorFlower
gardener@tensorflow.org
1,694,025,414,000
f9a7703009e788699d6d0a90179844f26bf18cc9
#tf-data-service Clean up an extra space for repetition logging. PiperOrigin-RevId: 563171180
Yang Chen
yangchen@google.com
1,694,026,129,000
a78a3b15ebcf1a19248b6c42a5f704b7911fbb31
[Launch] Use PJRT (Device API) for GPU compilation. PiperOrigin-RevId: 563172526
Swachhand Lokhande
swachhand@google.com
1,694,026,402,000
f48c61c71d7d1e2b065ea622921fcf81281a8a13
don't create (empty) power_metrics if doesn't exist. PiperOrigin-RevId: 563173001
Jie Sun
jiesun@google.com
1,694,026,496,000
132c98b29c1a42298cd13055d1168c8d67177bc9
Vendor XLA/TSL into TensorFlow as Bazel dependencies This marks an important step towards delivering OpenXLA, Google’s OSS-based unified ML software infrastructure. PiperOrigin-RevId: 563177047
Jake Harmon
jakeharmon@google.com
1,694,027,275,000
60050130d46d367816fdb5bf917bf0c2d3b688d1
Adds an explicit log statement when preserving XLA sharding annotations. PiperOrigin-RevId: 563190066
A. Unique TensorFlower
gardener@tensorflow.org
1,694,030,167,000
85784a4aa9cb9a918f17ae9dabb08edb9c444fa0
Drop tf/python/layers ref in tensorflow/python/trackable. PiperOrigin-RevId: 563193915
Adam Cogdell
adamcogdell@google.com
1,694,030,943,000
81658fc94bdb7ca0956cfc1a7907f1bd37e36c8a
Update configure.py to support NDK 25c
Terry Heo
terryheo@google.com
1,694,034,525,000
c121643b7e125891d9314631f3d6f0a772737c66
[XLA:CPU] Fix random crashes on Windows due to out-of-order RuntimeDyld sections. On Windows, the JAX test suite has random crashes that turn out to come from the LLVM fatal error "IMAGE_REL_AMD64_ADDR32NB relocation requires an ordered section layout", which originates from the LLVM runtime dyld code: https://github....
Peter Hawkins
phawkins@google.com
1,694,036,207,000
eea1fd48e749f12ee75f46d41b3fdfc4e2678673
Promote the TF Lite signature runner C APIs from experimental to stable. PiperOrigin-RevId: 563219029
Fergus Henderson
fergus@google.com
1,694,036,213,000
0fcab8bb66e37c6dce16f9e41556cd92b87372e6
Encoding binary sharding to human readable while import def Decoding back to binary from human readable before exporting graph def PiperOrigin-RevId: 563222170
Yiming Zhang
yimingz@google.com
1,694,036,900,000
86e13790f92f37bf9f47148c34af39e67852d115
Remove public-facing copybara hooks from new scripts These are not relevant or useful for OSS users, and they've been moved to the internal side of the CI system. There is no loss in functionality. PiperOrigin-RevId: 563226965
Austin Anderson
angerson@google.com
1,694,037,918,000
075650f5b215394b3e124ba269da53ff228fe15b
Simplify import of ml_dtypes in C++. We actually don't need the pywrap module if we import types directly from the python ml_dtypes package, both in python and C++ (via pybind11). Removed the custom pywrap_ml_dtypes library, and directly import `ml_dtypes` from python. PiperOrigin-RevId: 563233368
Antonio Sanchez
cantonios@google.com
1,694,039,310,000
8ac85c3dbf0b8c289c5a6c351a2f1a576ed187b7
Drop tf/python/layers ref in tensorflow/core/platform/ram_file_system_test.py. PiperOrigin-RevId: 563234065
Adam Cogdell
adamcogdell@google.com
1,694,039,483,000
c5d8202b7edfeef0699fb8f29e2656300e0d8b1f
[XLA] Add WithContractingDims in pattern matcher and modify tests to conform to the new pattern matching format -Add WithContractingDims implementation for HloInstructionPattern to match with the dot instruction's LHS and RHS contracting dimensions. PiperOrigin-RevId: 563248293
A. Unique TensorFlower
gardener@tensorflow.org
1,694,042,838,000
d650701db88d4ab29b9e42558653e538573fd622
Call `madvise` with `MADV_PAGEOUT` on the unpackaged 4 bit weights after packaging. This evicts unused pages of the model, reducing the resident memory size. It is ineffective on operating systems other than Linux (e.g. Windows, but also other POSIX systems), requires kernel version 5.4 or newer, and still leaves on ...
Robert David
lrdx@google.com
1,694,043,494,000
b548efac3a366242dfecbaa6a27ce497cf1b2729
Internal developments for tfrt. PiperOrigin-RevId: 563256968
A. Unique TensorFlower
gardener@tensorflow.org
1,694,045,115,000
88d6d54a2438b8ad454d2263846ed53fdf8de924
Fix type hints otherwise running pytype will complain. PiperOrigin-RevId: 563257263
A. Unique TensorFlower
gardener@tensorflow.org
1,694,045,185,000
99e597d2cf96a966acd9a6212686946368f85fcf
#tf-data Change a log level in RBM from ERROR to WARN It is not that unusual for the free memory to be zero when an iterator is initialized. PiperOrigin-RevId: 563265637
A. Unique TensorFlower
gardener@tensorflow.org
1,694,047,664,000
040540b313e0e57d910b280d72a8687f482b0cd7
Remove an extraneous TODO comment that is no longer applicable. PiperOrigin-RevId: 563274736
Fergus Henderson
fergus@google.com
1,694,050,452,000
fc431d832880a73c53f086e5f7a59960a2fe9507
Fix tf_kernel_library definition involving textual headers. Some of the existing shared object libraries have been broken for a while, including ``` //tensorflow/core/kernels:libtfkernel_cast_op.so ``` since textual_hdrs were excluded. Added an argument to `tf_kernel_library` to allow explicitly specifying textual he...
Antonio Sanchez
cantonios@google.com
1,694,052,837,000
127d95557568f8fad267a4fbe5281ce00aa54de0
Use more fine-grained dependencies to avoid unnecessary dependencies on flatbuffer_utils. In particular, depend on //third_party/tensorflow/lite/core/api:error_reporter rather than //third_party/tensorflow/lite/core/api:api PiperOrigin-RevId: 563282469
Fergus Henderson
fergus@google.com
1,694,053,364,000
e992af659f10c6de3c482183a0369c99c78b213a
Implements HistogramMaxFrequency calibration method. PiperOrigin-RevId: 563284354
Junho Kim
kimjunho@google.com
1,694,054,222,000
a17714ddda170b7f3232b817a9264207b8bac90e
Add `GetOutputElementTypes` and `GetOutputDimensions` impls to `PjRtCApiExecutable`. Also forbid direct use of `GetOutputShapes` in `PjRtCApiExecutable` and `PjRtCApiLoadedExecutable`. PiperOrigin-RevId: 563285213
A. Unique TensorFlower
gardener@tensorflow.org
1,694,054,534,000
28dba7e8d78b290803d5f7b2f7abd7f65e3d2503
Merge pull request #59939 from tensorflow:tiruk007-patch-3 PiperOrigin-RevId: 563294667
TensorFlower Gardener
gardener@tensorflow.org
1,694,058,100,000
142dd70874f42aedaadb5d7f3c74e25d6acf28c8
Merge pull request #61756 from tensorflow:pjpratik-patch-10 PiperOrigin-RevId: 563294892
TensorFlower Gardener
gardener@tensorflow.org
1,694,058,478,000
e47dd4bea15aef6af1a51f3363feba31e9bc73c1
Refactoring: Move test-related code out of runOnOperation This makes the boundary between production and test code clearer. PiperOrigin-RevId: 563297980
Thai Nguyen
thaink@google.com
1,694,058,953,000
823da868466b1bb0768325a79d25d1ad00b4fc9b
[XLA] MakeShape* should terminate on failure Those functions don't return a Status so just logging seems dangerous. PiperOrigin-RevId: 563307297
David Majnemer
majnemer@google.com
1,694,062,290,000
d30a736840e8aa515da3b9e3445890bc4ba4c4aa
Internal Code Change PiperOrigin-RevId: 563311202
A. Unique TensorFlower
gardener@tensorflow.org
1,694,063,476,000
1bf6db173af2b08d8dc5bf6b86875e1bf9378540
Add index annotation for pybind extension modules in saved_model. PiperOrigin-RevId: 563316254
Daniel Chen
chencm@google.com
1,694,064,882,000
95cb42ff6a211d6d65431d50554d92a7fc5d8392
[XLA] Clean up some bit-cast/type logic No functional changes. PiperOrigin-RevId: 563317332
David Majnemer
majnemer@google.com
1,694,065,200,000
16073f85f69cbffe56c7c2a0021d45af3e355711
[Linaro:ARM_CI] Remove mention of XLA unit tests The XLA code is no longer present in the TensorFlow tree so need to remove mention of the XLA unit tests from the skip list
Andrew Goodbody
andrew.goodbody@linaro.org
1,694,072,526,000
0e94762b366cfbd9e159b11ef29abbc8650b58bf
[XLA:GPU] Pass upper_bound to EstimateThreadCount. We can take the min inside the function, so it removes the need to look into the optional to check for the value. PiperOrigin-RevId: 563356742
Oleg Shyshkov
shyshkov@google.com
1,694,076,008,000
4dad017378d30f8ea59f79440efe0891767965e3
Update GraphDef version to 1612. PiperOrigin-RevId: 563360827
A. Unique TensorFlower
gardener@tensorflow.org
1,694,077,330,000
8dc155574a9c08aac4dd147f556baae58a7f785b
compat: Update forward compatibility horizon to 2023-09-07 PiperOrigin-RevId: 563360828
A. Unique TensorFlower
gardener@tensorflow.org
1,694,077,330,000
ae5cd8d10a423fe4e8ceed813e125b5869db2237
Add visibility of :resource_base for learning/brain/google/data/core/kernels:table_dataset_op. PiperOrigin-RevId: 563361074
A. Unique TensorFlower
gardener@tensorflow.org
1,694,077,374,000
5276bc70fbc6d1a430a4d58d1fca4d8d57e6e0b2
Implements HistogramSymmetric calibration method. PiperOrigin-RevId: 563369679
Junho Kim
kimjunho@google.com
1,694,080,001,000
c4240d1980d44cb513941bdb66dc180d6347053f
Update op_tests to run multiple cases for rsqrt
Saoirse Stewart
saoirse.stewart@arm.com
1,694,080,712,000
276425fa40eebaf27908b37e91f5c9d64c44369d
Added placeholder for internal RPC option. PiperOrigin-RevId: 563389862
A. Unique TensorFlower
gardener@tensorflow.org
1,694,086,246,000
2ae9db09adbafca4d394c5434c5194227e17203a
Set default values for `mhlo.convolution` in `legalize_hlo.cc` when not set PiperOrigin-RevId: 563414773
A. Unique TensorFlower
gardener@tensorflow.org
1,694,094,241,000
2cc0aaca4d569d42cc546d69d9132bddc703b88b
PR #5472: [ROCm] Fix rocm register, GraphNodeGetType build error and rocm_blas stream Imported from GitHub PR https://github.com/openxla/xla/pull/5472 this PR has fixed rocm build errors due to 1. RegisterTraceListener https://github.com/openxla/xla/commit/335c8a09aa4de15ac565b0eab2baa7dcc5882c0f 2. GraphNodeGetType...
Chao
cchen104@amd.com
1,694,094,487,000
aa5459496f15f723f2eb9dffab9c573c93b113c4
Integrate LLVM at llvm/llvm-project@4acc3ffbb0af Updates LLVM usage to match [4acc3ffbb0af](https://github.com/llvm/llvm-project/commit/4acc3ffbb0af) PiperOrigin-RevId: 563421376
Dmitri Gribenko
dmitrig@google.com
1,694,096,022,000
dd163ecdc49b12bb55252b0870d2e459b16fffb4
Define macro TFLITE_CONDITIONAL_NAMESPACE, for use in code using cc_library_with_tflite. PiperOrigin-RevId: 563446264
A. Unique TensorFlower
gardener@tensorflow.org
1,694,102,101,000
ccf1b0ae66d658e8ec5ec7e5f6dce91fa58c9f2d
Add a range check on the axis index in `xla::ifrt::ShardingParam::verify()`. PiperOrigin-RevId: 563451890
Hyeontaek Lim
hyeontaek@google.com
1,694,103,235,000
9d425422c507802f7de096bcd06cabe554ec9af1
Improve Lowering for mhlo::uniform_quantize op in ConvertMHLOQuantToInt PiperOrigin-RevId: 563454255
A. Unique TensorFlower
gardener@tensorflow.org
1,694,103,786,000
d0934398a903cb7a3a50be0e403a2d00ef95f9d0
Merge pull request #61807 from elfringham:openxla PiperOrigin-RevId: 563456151
TensorFlower Gardener
gardener@tensorflow.org
1,694,104,484,000
0ce0eba07544f4f92e71da0a0e590d381499ebc0
Fix include paths for ml_dtypes in pip package. They are now relative to a top-level ml_dtypes folder. PiperOrigin-RevId: 563456963
Antonio Sanchez
cantonios@google.com
1,694,104,371,000
437d834efedc44ada968c1f64920a42383b1d4ea
Improve Lowering for mhlo::uniform_quantize op for requantize in ConvertMHLOQuantToInt PiperOrigin-RevId: 563463694
A. Unique TensorFlower
gardener@tensorflow.org
1,694,105,835,000
30373daf64d79e3b4d9bd6ea0487fb2248fc12c8
Fix a bug in ConvertTFQuantTypes pass that may causes failures when some but not all operands/result of an UQ op has already CastOp added. PiperOrigin-RevId: 563464822
A. Unique TensorFlower
gardener@tensorflow.org
1,694,106,065,000
98bc9b6b02b5e634c2c996efb97b797d97646bd8
Fix type hints otherwise running pytype will complain. PiperOrigin-RevId: 563465142
A. Unique TensorFlower
gardener@tensorflow.org
1,694,106,128,000
b2de4889b90e3da558cf626af397396de33175c2
Optimize `_PforInput.expanddim_inputs_for_broadcast` for statically known ranks. PiperOrigin-RevId: 563467365
A. Unique TensorFlower
gardener@tensorflow.org
1,694,106,537,000
4673723c6269413e299afc23893aa5611a113ff6
Regenerate pyi file PiperOrigin-RevId: 563467446
A. Unique TensorFlower
gardener@tensorflow.org
1,694,106,552,000
e75603d74de2f9f0c869c77017df83c16b0c0498
Moves the model config stub out of the saved_model_lib so that it keeps the saved model size small. PiperOrigin-RevId: 563477312
A. Unique TensorFlower
gardener@tensorflow.org
1,694,108,361,000