hash stringlengths 40 40 | msg stringlengths 1 131k | author stringlengths 1 33 | email stringlengths 0 57 | date int64 1,447B 1,698B |
|---|---|---|---|---|
eee2f3d45ca5fb94ee58a6db3cb6c59d437cfcd2 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/271ffac9f8ea91a556d6f1eca727cab80d39f90a.
PiperOrigin-RevId: 571347063 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,606,814,000 |
163fc8bfba816b313b4f049987c6b7750039f86f | Split client target into single-source targets.
PiperOrigin-RevId: 571364191 | Juan Martinez Castellanos | juanantoniomc@google.com | 1,696,611,237,000 |
14889277cb2266863033ed400643b60f4bd900e9 | Skip TPU V1 Compat bridge for inference graph outside execution of TPUPartitionedCall op
PiperOrigin-RevId: 571364374 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,611,291,000 |
478ef7c6f8a349f20c74090092391a1f41f47ae7 | No public description
PiperOrigin-RevId: 571364597 | Jim Lin | jimlintw@google.com | 1,696,611,357,000 |
3b3a3946ccb5f8b847b5d98be21e31444947f3bf | Create CI env file for new nightly ARM64 experimental build
PiperOrigin-RevId: 571384702 | Michael Hudgins | michaelhudgins@google.com | 1,696,615,712,000 |
78fc05556420bdf40283342588a17b7ee13baf3e | Skip TPU nominal bridge for inference graph
PiperOrigin-RevId: 571388494 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,616,606,000 |
3d3d3af519c00f1bc937e10af050b2f3f6714aa1 | [XLA:GPU] Add a hook to enable uploading HLO to internal infrastructure.
PiperOrigin-RevId: 571389810 | pizzud | pizzud@google.com | 1,696,616,868,000 |
f36d7785e02d18fc23e8215d6e0fa4b02823042b | Add CI configs for macOS Arm64 builds
PiperOrigin-RevId: 571390990 | Nitin Srinivasan | srnitin@google.com | 1,696,617,127,000 |
b65cadbb23c37e88f3652859a19b35ba5190e114 | Skip test for failing kokoro mac test
PiperOrigin-RevId: 571392275 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,617,409,000 |
c6a0d64b990bbe4410052af66991ab9e0786eda9 | Add data to `tf_python_pybind_extension` and `tf_python_pybind_extension_opensource`.
PiperOrigin-RevId: 571393106 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,617,589,000 |
2ce1d3301da8d5276bfa42aa05b2a363d7b5aa33 | Make embedding preprocessing collective gathers stateless
PiperOrigin-RevId: 571397247 | Pat Notz | patn@google.com | 1,696,618,541,000 |
8ab9107a3c479bb5100a6f9b9572c1112b5e5b1a | [XLA] Fix minor typo.
PiperOrigin-RevId: 571397879 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,618,695,000 |
dbcfdeb64c31c483faebc952d82ceb70bfc48e4c | Add an environment variable PJRT_NPROC that may be used to set threadpool sizes.
If both PJRT_NPROC and NPROC are set, prefer PJRT_NPROC. This allows a PJRT-specific NPROC override.
PiperOrigin-RevId: 571400733 | Peter Hawkins | phawkins@google.com | 1,696,619,277,000 |
45003ffedf35e3c1534743c649d039d65597f6a5 | PR #56789: [iOS] Add initial xcframework bazel support
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/56789
This adds xcframework support for 2 of the core libraries now that the
public bazel rules support producing static framework based
xcframeworks. This forks the hide symbols script to be compatible with
operating on multiple frameworks inside a given xcframework. Currently
this rules feature requires bazel 6.x rolling releases because it uses a
new API from the C++ starlark migration. That release should be public
in the next few months, for now this doesn't affect 5.x because the new
targets are manual.
Copybara import of the project:
--
5e90cde3004bd8bc38ab8853f01dc4b2fbe91164 by Keith Smiley <keithbsmiley@gmail.com>:
[iOS] Add initial xcframework bazel support
This adds xcframework support for 2 of the core libraries now that the
public bazel rules support producing static framework based
xcframeworks. This forks the hide symbols script to be compatible with
operating on multiple frameworks inside a given xcframework. Currently
this rules feature requires bazel 6.x rolling releases because it uses a
new API from the C++ starlark migration. That release should be public
in the next few months, for now this doesn't affect 5.x because the new
targets are manual.
Merging this change closes #56789
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/tensorflow/pull/56789 from keith:ks/ios-add-initial-xcframework-bazel-support 5e90cde3004bd8bc38ab8853f01dc4b2fbe91164
PiperOrigin-RevId: 571405194 | Keith Smiley | keithbsmiley@gmail.com | 1,696,620,379,000 |
dc4d70f61f63942e048193cb2e250b02a12e20d6 | Propagate stack traces when creating a GraphExecutionState. Constructing a FunctionLibraryDefinition was previously throwing away stack traces. Extract them from the GraphDef, and store them in the FunctionLibraryDefinition.
PiperOrigin-RevId: 571415410 | Gabriel Rasskin | grasskin@google.com | 1,696,622,812,000 |
5428209b50c4750bd1df451d60e44ca386ce18da | For aarch64, use empty kTargetCpuForHost string
The string "aarch64" does not appear in the list of valid CPUs in
AArch64.td, making it not a valid processor in
llvm/lib/MC/MCSubtargetInfo.cpp:getFeatures, which reports an
"unrecognized CPU" error, ultimately hitting an llvm_unreachable
assertion during code lowering when running tests such as
cpu_eigen_dot_operation_test. Log of erroneous behavior below.
This is very similar to the x86_64 machine description, which also
does not have a true "x86_64" processor listed. Using the
empty string as the cpu type gets translated to "generic", which is
handled properly by the rest of the infra.
Match that behavior for aarch64, which allows the test to pass properly.
Failing log for cpu_eigen_dot_operation test when run on an aarch64 host:
"""
[==========] Running 6 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 6 tests from CpuEigenDotOperationTestInstantiation/CpuEigenDotOperationTest
[ RUN ] CpuEigenDotOperationTestInstantiation/CpuEigenDotOperationTest.SimpleDotOp/F16
'aarch64' is not a recognized processor for this target (ignoring processor)
'aarch64' is not a recognized processor for this target (ignoring processor)
'aarch64' is not a recognized processor for this target (ignoring processor)
'aarch64' is not a recognized processor for this target (ignoring processor)
'aarch64' is not a recognized processor for this target (ignoring processor)
'aarch64' is not a recognized processor for this target (ignoring processor)
Don't know how to custom expand this
UNREACHABLE executed at llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:24052!
"""
PiperOrigin-RevId: 571415910 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,622,937,000 |
8030c0de053a048abac102d3f61d15d1d3ef9962 | [xla] Remove the explicit dependence added from RecvDone to Send.
This dependence is correct for unpipelined Send and Recv sequences but not for
pipelined Send and Recv sequences. We now expect a backend to add control
dependence to represent the intended ordering of Send and Recv, as the GPU HLO
scheduler does, via P2PSchedulePreparation pass.
PiperOrigin-RevId: 571430708 | Bixia Zheng | bixia@google.com | 1,696,626,481,000 |
ff37a11772d149f19a8e8f487f5f521865dde0f2 | Add op for writing statistical information about minibatches, for FDO.
PiperOrigin-RevId: 571431478 | Matthias Kramm | kramm@google.com | 1,696,626,695,000 |
952b26c17dc2833107a96f6c22120fa3f4849621 | Split framework target into single-source targets.
PiperOrigin-RevId: 571433450 | Juan Martinez Castellanos | juanantoniomc@google.com | 1,696,627,195,000 |
c431e43017146a0c8f26ced15d82172445299cf8 | [xla] Make autotune_results_proto header only by default
PiperOrigin-RevId: 571434928 | Eugene Zhulenev | ezhulenev@google.com | 1,696,627,566,000 |
1b76764ce81fa20cd51ae9ca2172f0b0215fd22e | Emit an error message if we fail the embedding_program_key pass verification.
PiperOrigin-RevId: 571442001 | Matthias Kramm | kramm@google.com | 1,696,629,345,000 |
d3567a27fcc33de5f59208f8b345a97df7e11af3 | Allows `grappler.tf_optimizer.OptimizeGraph` to process `MetaGraphDef`s larger than 2 GiB.
With larger models becoming more prevalent, it is not inconceivable to run into `MetaGraphDef`s larger than 2 GiB. Protobufs larger than 2 GiB cannot be serialized, making these models impossible to process. This commit removes the `SerializeToString` function, and uses pybind to pass the `MetaGraphDef` protobuf from Python to C++. By default, pybind also serializes the protobuf when passing. Using `//third_party/pybind11_protobuf:native_proto_caster` configures pybind to not do any serialization.
PiperOrigin-RevId: 571446639 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,630,518,000 |
2674ecf8c851a8b4aafaf38537ea39fa70d82db0 | Create ProfilerStatusHelper or profiler.
PiperOrigin-RevId: 571451461 | Clive Verghese | cliveverghese@google.com | 1,696,631,793,000 |
a09d7535f5514131c4c9c935938f89f84a278d26 | Fix a typo in array_impl_test_lib: reshared -> resharded.
PiperOrigin-RevId: 571452967 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,632,228,000 |
48ae913f1bfe6207e9a44b5ff1fdbe67720c89ce | Handle the case when the operand to an optimziation barrier HLO is a tuple.
PiperOrigin-RevId: 571464553 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,635,650,000 |
e4485c98eaed317a7648ad0f69f4fa170237b967 | #tf-data-service Add a separate to the temp files to make them easier to parse.
PiperOrigin-RevId: 571466104 | Yang Chen | yangchen@google.com | 1,696,636,120,000 |
0408e4e42af80be6ed95bf416e17b7b1a6be1ec9 | Avoid UB when casting from float to int. casting ULONG_MAX to float creates a value that is outside of the range of unsigned long (round to nearest). Casting that to unsigned long results in UB.
PiperOrigin-RevId: 571508474 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,654,387,000 |
78de18d6e20e6d6dd2a9716d168d950655f73cda | [XLA:GPU] Remove some Windows- and Darwinn-specific code.
PiperOrigin-RevId: 571523075 | Christian Sigg | csigg@google.com | 1,696,662,209,000 |
333145db599786dc253c26ad87c83201afd4811a | [hlo_evaluator] Don't create blank, zero-populated Literals for tuples
When evaluating Tuple(), create a blank Literal for the destination that is populated with
kUndetermined.
We already copy in all known leaves, so all this does is avoid us creating blank leaf Pieces,
which is incredibly important when dealing with large programs pre-sharding. Also note that
we'd instantaneously throw the blank pieces all away...
PiperOrigin-RevId: 571532410 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,666,635,000 |
a47d18e3f3c016957b1dacf6052d07aa606b7022 | [TSL] Remove some Windows- and Darwinn-specific GPU code.
PiperOrigin-RevId: 571535260 | Christian Sigg | csigg@google.com | 1,696,668,122,000 |
5dd7a1ef027418a8ecb07c93c62a2bd264e09258 | compat: Update forward compatibility horizon to 2023-10-07
PiperOrigin-RevId: 571537774 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,669,315,000 |
deec1654830f79e210856048487b31e23727c327 | Update GraphDef version to 1642.
PiperOrigin-RevId: 571537775 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,669,315,000 |
233ad03685295c15f99af44bbf516d4877b1382e | Add transpose to the mhlo->tfl path
PiperOrigin-RevId: 571593152 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,698,597,000 |
34821bdb9657be3c3253083cf8304b345f515fb4 | Use tf2xla implementation of Concat_v2 op for MLIR bridge
PiperOrigin-RevId: 571598203 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,700,764,000 |
4039065c9b12191326ac025a0a8d09ac3abbf41f | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/e8f4445c1129af7356c0a0c0a4ca60664b150d97.
PiperOrigin-RevId: 571669755 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,740,449,000 |
9c4a2d1c66822ae6b8a025075170a3d5ce2379de | Update GraphDef version to 1643.
PiperOrigin-RevId: 571701196 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,755,738,000 |
b72d2fd925012d578dcb3c2da9970525791e051a | compat: Update forward compatibility horizon to 2023-10-08
PiperOrigin-RevId: 571701204 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,755,740,000 |
fc62152373da52f9631b4aa566ebc9144999cda0 | Fix ReduceWindow kernel output shape computation.
PiperOrigin-RevId: 571754187 | Quentin Khan | qkhan@google.com | 1,696,786,627,000 |
75366c399669201c541f028e3127bc5bcb618a75 | Runs requirements updater to fix nightly Python 3.11 error.
Error is caused by pypi_wrapt hash mismatch. See https://github.com/tensorflow/tflite-micro/issues/2264.
PiperOrigin-RevId: 571775303 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,799,497,000 |
75d8f13cbf1946a3a929cb4c89e26be964044a27 | Merge pull request #61738 from Tai78641:pr_fix_broadcast_to_lowering
PiperOrigin-RevId: 571820803 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,824,520,000 |
fe9cdf5d8b68298908a4eb52ffbdf635e75c61aa | Merge pull request #61941 from ROCmSoftwarePlatform:hipblas_lt_integration2
PiperOrigin-RevId: 571822200 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,825,234,000 |
34660887138690c0186304ee9298dc1042583c8c | Merge pull request #61998 from tensorflow:tilakrayal-patch-1
PiperOrigin-RevId: 571823284 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,825,832,000 |
abbbcb50208cf6d5a965b7bf60efec5bb078e142 | Merge pull request #61990 from Tessil:reshape_with_dynamic_dims_fix
PiperOrigin-RevId: 571824268 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,826,395,000 |
df0af2cd9ca6fed4d1110b5186b6b9bc4efa5a58 | Internal Code Change
PiperOrigin-RevId: 571866788 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,840,400,000 |
9c17ff9d1c05e99ca3227b6b6468ef2371f7f738 | Integrate LLVM at llvm/llvm-project@2a1f1b5fde0a
Updates LLVM usage to match
[2a1f1b5fde0a](https://github.com/llvm/llvm-project/commit/2a1f1b5fde0a)
PiperOrigin-RevId: 571867638 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,840,731,000 |
4167d2eea8a3c4355fb3cd01d1b43f74136c2c89 | PR #6150: Correction of CudaAsyncAllocator, not BFCAllocator
Imported from GitHub PR https://github.com/openxla/xla/pull/6150
Fix the comments and logs for CudaAsyncAllocator, not BFCAllocator which is a synchronized cuda memory allocator.
Copybara import of the project:
--
5cfd652cc759c28d6f03d42e82d92bf19d0c558a by Jane Liu <janeliu@nvidia.com>:
Correction of CudaAsyncAllocator, not the BFCAllocator
Merging this change closes #6150
PiperOrigin-RevId: 571868157 | Jane Liu | janeliu@nvidia.com | 1,696,840,907,000 |
526851e85d0ee9198bcf46a9badee1035f162144 | Update GraphDef version to 1644.
PiperOrigin-RevId: 571871679 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,842,118,000 |
93774d54c4d44d96f59d56e56a5c6ef10b4f1ac5 | compat: Update forward compatibility horizon to 2023-10-09
PiperOrigin-RevId: 571871706 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,842,124,000 |
a859444925752d7a98a3747b14ca96d2987d4420 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/30adf6e26966b1b3413447e8922f450303238dc5.
PiperOrigin-RevId: 571874457 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,842,899,000 |
183ea613917ff133888bbb47e9052bef2e133eb6 | Internal Code Change
PiperOrigin-RevId: 571891742 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,848,379,000 |
f91457f258fdd78f693044a57efa63a38335d1de | Update rules_python: 0.25.0 -> 0.26.0.
PiperOrigin-RevId: 571910358 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,854,486,000 |
ae5d695b8a25a593a424169159bbd07cff3fa728 | Add gpu_compiler dependency inside if_cuda_or_rocm block.
This avoids a duplicate dependency problem if both if_cuda and if_rocm blocks
are taken.
PiperOrigin-RevId: 571911265 | Adrian Kuegel | akuegel@google.com | 1,696,854,833,000 |
3c406adebe6e8cb7b1a70d79b5cb6a2b45fd57ec | Moves the tensor dimension enumeration logic out of the individual split methods.
PiperOrigin-RevId: 571959782 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,869,087,000 |
aeceebbfa297c26c2319e5c2b754ffc7a8ac90be | Sprawling .pyi updates related to pybind11 PRs #4831, #4833.
PiperOrigin-RevId: 571968911 | Ralf W. Grosse-Kunstleve | rwgk@google.com | 1,696,871,052,000 |
4a5ba89cea0c26028dba375a338873ea17d430e4 | Eliminates the awkward Split([this](MeshDims md) { this->SplitXYZ(md); }) stuff by making the Enumerate() method take care of the mesh dimensions.
PiperOrigin-RevId: 571969979 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,871,274,000 |
58608f98f49f12d25bd4a9f28efd75f7f00f8dfd | Move V1CompatBridge into ClusterTF and migrate users to call this API.
PiperOrigin-RevId: 571970849 | Mason Chang | masonchang@google.com | 1,696,871,436,000 |
3cccb1ba0e80fd62cb644bd87baeec86141568be | Re-enable layering checks.
PiperOrigin-RevId: 571974004 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,872,130,000 |
0f591ff0bbc2ee8ee764d190fc27e10fde9acfa8 | Increase minimum scipy version to 1.9.
Scipy 1.9 appears to fix some crashes on Mac ARM.
PiperOrigin-RevId: 571977068 | Peter Hawkins | phawkins@google.com | 1,696,872,792,000 |
27509039b1d54decaf86880fe64d1f31195d8abb | Split lib target into single-source targets.
PiperOrigin-RevId: 571982368 | Juan Martinez Castellanos | juanantoniomc@google.com | 1,696,873,816,000 |
3d16828c2223ee9ee8660a0cadfa986c2182505d | Consider instructions with 1 user in HloConstantSplitter.
With this cl, HloConstantSplitter may generate dead instructions. HloDCE is added after it in the realted compiler pipelines.
PiperOrigin-RevId: 571999001 | Zixuan Jiang | zixuanjiang@google.com | 1,696,877,269,000 |
5ec41a3eba638b65e3f7325a583ecefa9e43a710 | #tf-data-service Clean up snapshot repetition logging.
PiperOrigin-RevId: 571999215 | Matt Callanan | mpcallanan@google.com | 1,696,877,321,000 |
23d5e1da23d2e828a91c9cabf0ebad5c744a955a | Redirect references from the framework target to the new single-source-file targets.
PiperOrigin-RevId: 572000652 | Juan Martinez Castellanos | juanantoniomc@google.com | 1,696,877,616,000 |
ec712ed2e9a8e88234ba377f27120eaebb28d3f0 | Add `TF_ResetPjRtCClient` to remove a pjrt client from the state
PiperOrigin-RevId: 572000778 | Haibo Huang | hhb@google.com | 1,696,877,636,000 |
e0df77c2793358755bdeddf2f76442845311fcde | Handle a case in FollowReduceStrategy where we do not wish to crash if trying multiple mesh shapes. Also bump up the VLOG level of a module dump before auto-sharding as usually we have HLO dumps on, and the modules dumped are huge leading to large log files that become difficult to navigate.
PiperOrigin-RevId: 572006217 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,878,772,000 |
73632cf3b799744c355014d1143500904e8c0e68 | Fix libtpu path in whl
This fixes 2 issues in the code:
1. `std::filesystem::canonical` requires the path to exist. Which is not always the case.
2. use after scope error in the code. `c_str()` will be invalid after the path object is destructed.
PiperOrigin-RevId: 572006731 | Haibo Huang | hhb@google.com | 1,696,878,888,000 |
f4143a40fe9063227ae95b5df7e7afb6a420f726 | [xla:gpu] Construct HloComputation from HloInstructionSequence
https://github.com/openxla/xla/issues/5756
PiperOrigin-RevId: 572008403 | Anlun Xu | anlunx@google.com | 1,696,879,284,000 |
ae308a6e99f6762549438af8542530387333a87e | Redirect references from the framework target to the new single-source-file targets.
PiperOrigin-RevId: 572012180 | Juan Martinez Castellanos | juanantoniomc@google.com | 1,696,880,241,000 |
e6341f821cc2a2e6022d31df7796d67dfb57b013 | [stream_executor] NFC: Add explicit command buffer state enum #5857
PiperOrigin-RevId: 572013357 | Eugene Zhulenev | ezhulenev@google.com | 1,696,880,519,000 |
8afcb090a8bb1210a16fc52b862ad0d0cb4c7617 | Legalize TensorListPopBack to tfl custom.
PiperOrigin-RevId: 572018125 | Luke Boyer | lukeboyer@google.com | 1,696,881,685,000 |
a58a7e3a0ad2a1dd970683eaee5f33a59f4bb80c | Add loop unrolling pass to the compiler.
This would allow the loops to get unrolled after all the hlo optimizations before buffer assignment.
PiperOrigin-RevId: 572023779 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,882,964,000 |
1e89b726f0d46cc53c1b69e82a3c332bcd87f85b | [PJRT C API] Adding Profiler C APIs and related framework changes.
C API changes:
- Profiler C APIs are added in profiler_c_api.h.
- Add a PJRT C API extension for the profiler C APIs in pjrt_c_api_profiler_extension.h.
Framework changes:
- Add a plugin_tracer that calls profiler C APIs.
- Add a pybind method xla_client.profiler.register_plugin_profiler to register plugin_tracer with the plugin's PJRT_Api*.
- Update xla_bridge.register_plugin to call register_plugin_profiler to register profiler for that plugin.
PiperOrigin-RevId: 572027222 | Jieying Luo | jieying@google.com | 1,696,883,733,000 |
467d3213d3ef35e49e2f9a06458691685cbfb3ba | [JAX] Update Python patch versions in RBE images used by JAX and enable compilation parallelism.
* Enable parallel compilation by passing `-j$(nproc)` to `make`
* Disable the `--enable-optimizations` flag to CPython, which controls profile-guided optimization, because it appears to be incompatible with parallel compilation under Python 3.11 (https://docs.python.org/3/using/configure.html#performance-options). We don't particularly care about profile-guided optimizations for the CI use case anyway: we aren't going to be particularly Python performance bound for JAX.
* Update the Python patch versions because there's no reason to use older buggy patch releases.
PiperOrigin-RevId: 572029030 | Peter Hawkins | phawkins@google.com | 1,696,884,112,000 |
ab10c2f8b37201a8413e790bb306a69706096cb2 | [stream_executor] Add PlatformSpecificHandle to Stream #6165
- Add API to access platform-specific details for a SE Stream
- Remove StreamExecutor internal headers and APIs from PjRt
`stream_executor_internal.h` header should not be accessible outside of SE, this change moves codebase closer to desired state.
PiperOrigin-RevId: 572029396 | Eugene Zhulenev | ezhulenev@google.com | 1,696,884,206,000 |
f6ef48b21ddeead791a005ef1bfabfa05b26add4 | Updates Auto Sharding to convert the liveness set into node indices only once (instead of once per solve).
PiperOrigin-RevId: 572032671 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,884,922,000 |
28e2df3e76427eace38e85ce7ea7052a49068ae0 | Check for unranked shape spec in FromTensor
PiperOrigin-RevId: 572033297 | Luke Boyer | lukeboyer@google.com | 1,696,885,080,000 |
02504cebf313ef09b5f2ffa450ffe19da7bc75ce | [XLA:LatencyHidingScheduler] Remove the redundant config.
This config is neither set nor used.
(Only scheduler_core_.config is used when needed.)
PiperOrigin-RevId: 572034182 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,885,267,000 |
8175a2aaf8cad1e2e4b52b4b2ae2ea987fb307a2 | #tf-data-service Add a test for repetitions with multiple workers.
PiperOrigin-RevId: 572042608 | Yang Chen | yangchen@google.com | 1,696,887,143,000 |
83078cb3cc4da3f2224e2e09825d1f164ccc2da5 | [gpu] NFC: Do not use StreamExecutor internal API to access underlying Gpu stream
PiperOrigin-RevId: 572045830 | Eugene Zhulenev | ezhulenev@google.com | 1,696,887,898,000 |
36feeda5b9786fec5ccd40bf036e085c0ba865ac | Clean up npd c_api.cc
As the API implementation, it can access TF_Status internals.
PiperOrigin-RevId: 572059363 | Haibo Huang | hhb@google.com | 1,696,891,095,000 |
3c7cb979048e9e7a8ddeea903d089c6e6546d2e6 | Set layering_check to ON by default and fixed targets with no visibility issues.
PiperOrigin-RevId: 572063126 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,892,004,000 |
9834ba6a29273804fab1fa8145c27b682830e87f | #tf-data Turn down `"file_locality"` (v1) experiment.
PiperOrigin-RevId: 572063544 | Matt Callanan | mpcallanan@google.com | 1,696,892,098,000 |
decc6300b49f14553e29c1b254708d2591339afa | Update logging instructions and group logging into appropriate clusters.
PiperOrigin-RevId: 572071735 | Mason Chang | masonchang@google.com | 1,696,894,271,000 |
c71ecec8421fb07c8680e638419d75a9803b94ca | Transform batch_matmul to fully_connected in tflite if the weight is rank2 constant.
PiperOrigin-RevId: 572072244 | Yishuang Pang | ypang@google.com | 1,696,894,393,000 |
e880e4476b1e79b30261dd2a310553d69229428d | Fix tf_registration_test by calling tf.saved_model.save to load the necessary registration.
PiperOrigin-RevId: 572075663 | Fiona Lang | flang@google.com | 1,696,895,308,000 |
f94012e91c2f7d7cc75fb0c160f8b4282a923dfd | handle some cases where the mlir and TF op statefulness is not consistent in the if op creation logic.
PiperOrigin-RevId: 572082936 | Ziyin Huang | ziyinh@google.com | 1,696,897,256,000 |
82344a1bad6a57cd2a3ef40867fa84f53d7d427b | Integrate LLVM at llvm/llvm-project@42c564df2f47
Updates LLVM usage to match
[42c564df2f47](https://github.com/llvm/llvm-project/commit/42c564df2f47)
PiperOrigin-RevId: 572084886 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,897,792,000 |
c534c2a95ce7f3e953080a61382da47edf24c283 | Record `enable_large_batch_splitting` as TF does
PiperOrigin-RevId: 572085950 | Juanli Shen | juanlishen@google.com | 1,696,898,127,000 |
db49583f9d02e449c1bdd4ea3809a4a97dd11e58 | Reuse TensorListSetItem to implement TensorListPushBack
PiperOrigin-RevId: 572087048 | Luke Boyer | lukeboyer@google.com | 1,696,898,423,000 |
51b64abf8c4e8b2b0153c04eadac6f524ad4683a | Record free GPU memory in PJRT.
PiperOrigin-RevId: 572087910 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,898,697,000 |
766edb75c0afddac6f26c5611c6d0bc08cbf84c0 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/70637966e2ec9afccc2cf4d51ed2391172b1b9c5.
PiperOrigin-RevId: 572093149 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,900,370,000 |
77c5a33394cf436a5de7a0babb47712a71364b39 | [stream_executor] Add support for updating primary command buffers (only launches for now) #5857
PiperOrigin-RevId: 572093629 | Eugene Zhulenev | ezhulenev@google.com | 1,696,900,550,000 |
31657c1bd93649cd226ef4155dcb2539d8b528fe | Revert initialize_fn param to fix Seg fault
PiperOrigin-RevId: 572094782 | Chandra Devarakonda | chandrasekhard@google.com | 1,696,900,956,000 |
1604ea054f81dbee65c793db5519ead5fc756df2 | #tf-data-service Add log for worker repetitions.
PiperOrigin-RevId: 572096819 | Yang Chen | yangchen@google.com | 1,696,901,607,000 |
dc4fbf0408e770898f82b0e6c6852d9693413bc5 | Merge pull request #61871 from Intel-tensorflow:mraunak/Clang_TF_Win
PiperOrigin-RevId: 572096891 | TensorFlower Gardener | gardener@tensorflow.org | 1,696,902,366,000 |
f250fabd3bcf4953db7fb17ca60e8520aa464b0d | Add support for StableHLO custom_call.
Legalizes `mhlo.custom_call` to `tfl.custom`. Not all facilities of the former
are supported. In particular, there is no support for `called_computations`.
Additionally, op specified with `backend_config` of type `mlir::DictionaryAttr`
will not be legalized, since the `tfl.custom` op doesn't have equivalent
support.
PiperOrigin-RevId: 572106685 | Arian Arfaian | aarfaian@google.com | 1,696,905,325,000 |
a46c83ec0b83bdc0d7b5ea3ce7e8e72250ae06b4 | Clear out invalid shardings from previous partial mesh runs before invoking sharding propagation. In some cases, sharding propagation can look at these shardings, thereby leading to invalid inferred shardings for operands.
PiperOrigin-RevId: 572114582 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,907,852,000 |
0ffca90a09afd54e6e0b152864503d4492a3c91a | #tf-data-service Add more tests for repeated dataset when workers are preempted.
PiperOrigin-RevId: 572115475 | Yang Chen | yangchen@google.com | 1,696,908,210,000 |
ee814127bfdf41c804660865847d53eccdf4e5a8 | [XLA:LatencyHidingScheduler] Improve the kDecreaseMemory rule by letting it take place only when it actually decreases the memory pressure.
Otherwise, it always picks the smaller-memory-increase HLOs, which does not necessarily obey the original order (which was already optimized by the memory scheduler). Picking the smaller-memory-increase HLO delays scheduling the larger HLOs further and further, so it increases the likelihood of getting stuck in local optima.
PiperOrigin-RevId: 572119189 | A. Unique TensorFlower | gardener@tensorflow.org | 1,696,909,503,000 |
188a263405532efb674c081799db136c75ff287f | Unify helper functions into TF path
PiperOrigin-RevId: 572123200 | Juanli Shen | juanlishen@google.com | 1,696,911,115,000 |
1d90751489c325cea7057712dcb7f7f1155a216e | Open source tpu_embedding_v3.py
PiperOrigin-RevId: 572131919 | Hye Soo Yang | hyey@google.com | 1,696,914,104,000 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.