hash stringlengths 40 40 | msg stringlengths 1 131k | author stringlengths 1 33 | email stringlengths 0 57 | date int64 1,447B 1,698B |
|---|---|---|---|---|
e54c8969d130799a6d2411ce7dd7d3d6551d934f | PR #5634: [ROCm] Fixed plugin config error
Imported from GitHub PR https://github.com/openxla/xla/pull/5634
Fixed ROCm build error due to https://github.com/openxla/xla/commit/7be97ae6ea0915136bbbe541bccb9e70f4c74248 forgot to include ROCm change.
Thanks in advance! @tdanyluk @cheshire
Copybara import of the project... | Chao | cchen104@amd.com | 1,694,802,542,000 |
9e2c09a295c3ea2a9adccc7627314147f3541c8b | Forward `experimental_attributes` to wrapped function creation calls.
PiperOrigin-RevId: 565736950 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,803,444,000 |
bc98a31f7975a1cdb6e289685124c06a19560c73 | oneDNN default on Windows | Raunak | mayank.kumar.raunak@intel.com | 1,694,804,189,000 |
b7fc45f1437d29b1522a815615f7457dd7e50afa | [xla:gpu] GraphExecUpdateResultInfo should be initialized to 0
So that it won't contain garbage values.
PiperOrigin-RevId: 565738255 | Anlun Xu | anlunx@google.com | 1,694,803,737,000 |
6acbb27ee9211d8214936510e05963e689b82af2 | Use different namespace for each .aar
Android requires all libraries to have a unique namespace defined in their manifest:
- Added manifests for the api and gpu libraries
- Renamed AndroidManifestGpu.xml to AndroidManifestGpuApi.xml for consistent naming | Bianca van Schaik | bvschaik@gmail.com | 1,694,803,945,000 |
de701b17dc2f1e83b17224b421e406dda38cdb8f | Includes `cstdint` in rng_util header.
PiperOrigin-RevId: 565741755 | Yishuang Pang | ypang@google.com | 1,694,804,489,000 |
4905187e03ed996b9e74b25515e35e2b72b9635c | #tf-data-service Do not acquire locks when writing splits.
PiperOrigin-RevId: 565744967 | Yang Chen | yangchen@google.com | 1,694,805,157,000 |
15363a2393bd54b1ed072af85940acf57ea48d9f | [mlir][sparse][xla] Legalize sparse_tensor::Pack/UnpackOp to custom calls before the translation to HLO.
PiperOrigin-RevId: 565749097 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,806,118,000 |
6107baae3ad1814811fc25b510f4ac45c478018c | #tf-data-service Instrument dispatcher RPCs.
PiperOrigin-RevId: 565751908 | Yang Chen | yangchen@google.com | 1,694,806,789,000 |
100b23067d7556bf3339585f5154fdcd9e7a9fb9 | #tf-data-service Add log for snapshot timing.
PiperOrigin-RevId: 565754578 | Yang Chen | yangchen@google.com | 1,694,807,440,000 |
5205391a12c430a9d93f006cf8879d16c39b941a | Fixed a bug that allocation block has 0 collocation when allocate_reserved_scoped_memory_at_same_offset is set false. This will cause minimalloc and telamalloc repacker to fail.
PiperOrigin-RevId: 565763337 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,809,589,000 |
6f41a94e653cbd0308c3f8619ae64c8bde2dd502 | [xla:gpu] Add runtime3 folder in preparation for runtime consolidation
PiperOrigin-RevId: 565768928 | Eugene Zhulenev | ezhulenev@google.com | 1,694,810,950,000 |
d1533803bc36d6a58db5d1ee8622b45022c7642e | Merge pull request #61300 from jamwar01:fix_tosa_rsqrt_table_diff
PiperOrigin-RevId: 565772533 | TensorFlower Gardener | gardener@tensorflow.org | 1,694,812,269,000 |
9b52021eca7bfdc96d1d485cec42f35e8fb72bd0 | Internal changes only.
PiperOrigin-RevId: 565786333 | Fiona Lang | flang@google.com | 1,694,814,988,000 |
723c41936274846907e31cd70018065f1f0db369 | Deduplicate envs by sourcing common settings
PiperOrigin-RevId: 565790027 | Austin Anderson | angerson@google.com | 1,694,815,925,000 |
3f0f23492b932cabe74317fd5be5d79a5cbc0bdd | Delete deprecated gpu tf_runtime backend
PiperOrigin-RevId: 565792209 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,816,514,000 |
91d784d8eba414e0c705937b6cd260c913304e08 | Refactor `TfrtCpuAsyncHostToDeviceTransferManager` by adding a new intermediate class `AbstractAsyncHostToHostMemoryTransferManager` between `PjRtClient::AsyncHostToDeviceTransferManager` and `TfrtCpuAsyncHostToDeviceTransferManager`. Other clients may use this new class as a shared impl to create their own async host ... | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,818,248,000 |
498397da44cfe303f50f0f9c5161731e6763a8cf | Temporarily disable host memory offload of tensors which have a tuple as a direct user.
PiperOrigin-RevId: 565802826 | Victor Stone | victorstone@google.com | 1,694,819,413,000 |
18ead88de74070414ae32333fd35e315f9d83d7a | Update `HloBufferDonorConfig::Verify` and `HloVerifier`. The overlap between `buffer_donor_config` and `input_output_alias_config` is not allowed.
PiperOrigin-RevId: 565803412 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,819,586,000 |
9235bf8f05a2677699aad0d3717eed4b40ecb625 | x86 platform included | Raunak | mayank.kumar.raunak@intel.com | 1,694,822,826,000 |
13baa8e7e50ca5ba2c95c12a6d686e7fa2596887 | Allow serializing env_options_overrides separately from the rest of
CompileOptions.
PiperOrigin-RevId: 565813066 | Parker Schuh | parkers@google.com | 1,694,822,572,000 |
c04d03db884defc1d3bf60e1b1b065b079c9593c | Use sharding propagation when possible to obtain a default solution to compare with the auto-sharding solution.
PiperOrigin-RevId: 565817073 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,823,976,000 |
162463fd8cd25b636ec42ed7618e5436f075369b | Replicate small constants so they don't need to be sent to their successors. A small constant is replicated to each of its successors' devices. The maximum size of a constant to be replicated is 16 elements.
This pass is disabled by default and can be enabled with the flag replicate_small_constants.
PiperOrigin-RevId... | Michael Delorimier | mdel@google.com | 1,694,825,451,000 |
4aac3064f48a1134e32bd8376ad90b0236034ecd | Open source op & kernel for `GetMinibatchSplitsWithPhysicalReplica`
PiperOrigin-RevId: 565837005 | Hye Soo Yang | hyey@google.com | 1,694,832,128,000 |
9297aa04f1c7384cd9d3540ab6e6877b72742117 | Update GraphDef version to 1621.
PiperOrigin-RevId: 565896297 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,855,076,000 |
fbe41b7246d137e99b892bbf2c3b54e9c615cec3 | compat: Update forward compatibility horizon to 2023-09-16
PiperOrigin-RevId: 565896351 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,855,089,000 |
1dc1e3cce02b3c06ffceb5a2f44e6e7ac92ec1f8 | [XLA] Make sure the dynamic sizes in literal are well aligned
This ensures that we do not trigger undefined behavior when accessing them.
PiperOrigin-RevId: 565980525 | David Majnemer | majnemer@google.com | 1,694,902,032,000 |
c212d26937b0ecde1400997c80a94deaa59ae9a2 | Resolved Comments | Raunak | mayank.kumar.raunak@intel.com | 1,694,912,018,000 |
7e00c4770cf6e29c00a09f6482bc784fc915e668 | added clang-cl build | Raunak | mayank.kumar.raunak@intel.com | 1,694,920,172,000 |
02ea5d1ba366015e712343e0f57fc6ea3766b382 | Fix Alignment | mraunak | 83710963+mraunak@users.noreply.github.com | 1,694,920,930,000 |
17d8892757be72019201e79981d311fd2a12b31e | Strip `external/local_tsl` prefix during tar of tsl c headers | William Muir | wamuir@gmail.com | 1,694,922,454,000 |
78a02bf38afdd008d5f3f03d041013a07c879c5d | compat: Update forward compatibility horizon to 2023-09-17
PiperOrigin-RevId: 566056330 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,941,325,000 |
a2fe2d61a52912e12c1b4f9d8b1834a7107c9a86 | Update GraphDef version to 1622.
PiperOrigin-RevId: 566056331 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,941,325,000 |
7a7ecd49dfabb12204d7fcbfebd00147542b27be | Fix deadlock in AbstractAsyncHostToHostMemoryTransferManager
PiperOrigin-RevId: 566119193 | A. Unique TensorFlower | gardener@tensorflow.org | 1,694,977,761,000 |
b2ec3bbcf93778a267d4a6c7bb45aed71b9d2546 | [pjrt] Add support for ARM Neon to transpose
The recursive transpose algorithm is pretty fundamental. We can implement it on Neon by just implementing some primitives.
While we are here, reduce code bloat by skipping instantiation of unspecialized micro-kernels.
PiperOrigin-RevId: 566137150 | David Majnemer | majnemer@google.com | 1,694,988,962,000 |
e4a6720f42adaa0adeabe04abbb7c6cf444329f9 | clean up the sparse core preprocess ops kernel
PiperOrigin-RevId: 566172603 | Ziyin Huang | ziyinh@google.com | 1,695,007,743,000 |
da646f1b1b9217b7e1694ae4b38caacb2335d090 | lite:stablehlo:transforms: Add comments for better readability.
PiperOrigin-RevId: 566194710 | Jiyoun (Jen) Ha | jiyounha@google.com | 1,695,015,823,000 |
5fd22c15c2a8a4152692ac442af394477e4d3e64 | (1/N) Refactor stablehlo pass namespaces to mlir::quant::stablehlo.
PiperOrigin-RevId: 566217901 | Jiyoun (Jen) Ha | jiyounha@google.com | 1,695,022,795,000 |
424006ba48b7894e9a1c908ae4e5d37b4927cf94 | [XLA::GPU] Do not keep TargetMachine alive after finishing executable compilation
This saves host RAM during compilation.
PiperOrigin-RevId: 566225133 | Tamás Danyluk | tdanyluk@google.com | 1,695,024,952,000 |
5b93ae202b214bb5eacc6cd56c09935fb6df7a60 | [XLA:GPU][NFC] Remove unused allocator argument from conv autotuning
PiperOrigin-RevId: 566228813 | Son Tuan Vu | vuson@google.com | 1,695,026,191,000 |
25f87b72ba7593966bf86aaf7cb8c7979aed6771 | [XLA:GPU][NFC] Add CreateMlirModuleOp as a leak-safe alternative of mlir::ModuleOp::create
mlir::ModuleOp::create returns a non-owning reference which is almost never the intended usage. We may leak memory if we don't assign it manually to an mlir::OwningOpRef. We actually had an error like this a few weeks ago.
Crea... | Tamás Danyluk | tdanyluk@google.com | 1,695,027,054,000 |
d9b1d2606107941a2f19cd16279f695f98875eca | Update GraphDef version to 1623.
PiperOrigin-RevId: 566233510 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,027,720,000 |
505c9f544caf2de04afd7f507c044896e5587994 | compat: Update forward compatibility horizon to 2023-09-18
PiperOrigin-RevId: 566233520 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,027,722,000 |
7d381bd7c680fc7f8da9fec0ccb1e6583ba282bc | [XLA:GPU] Fix build of HLO op profiler tool.
Add missing return statement, fix access to device name.
PiperOrigin-RevId: 566253614 | Ilia Sergachev | sergachev@google.com | 1,695,034,174,000 |
edd09bcc365dcc696d0f23ca7c3dc18f5e1dcdab | Check error code after quantizing
PiperOrigin-RevId: 566270089 | Alan Kelly | alankelly@google.com | 1,695,039,354,000 |
f6e9b5437803ad8a685f21f739bc3c655ae7dde0 | Integrate LLVM at llvm/llvm-project@79e96b2457fe
Updates LLVM usage to match
[79e96b2457fe](https://github.com/llvm/llvm-project/commit/79e96b2457fe)
PiperOrigin-RevId: 566274600 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,040,879,000 |
ac60f49d69557f9f2227ad84aea7a5f6c55d2845 | [XLA:GPU] Triton GEMM: support slicing of inputs.
With this logic slices will be fused quite rarely (only as inputs, either slices of direct computation parameters or tiny ones) because generally slices are better to fuse to producers to reduce DRAM traffic.
PiperOrigin-RevId: 566274725 | Ilia Sergachev | sergachev@google.com | 1,695,040,908,000 |
bc6d7a48432f1c7e1fd9b34ee3e90c919982ab8c | Fix atomic max-reduce edge case (negative zero)
PiperOrigin-RevId: 566278326 | Sergey Kozub | sergeykozub@google.com | 1,695,042,159,000 |
7805d33bf5ae7b184a4faaba02859689bd0f23d6 | PR #5670: [ROCm] revert 48cf922
Imported from GitHub PR https://github.com/openxla/xla/pull/5670
revert ROCm side same as https://github.com/openxla/xla/commit/5eb773450568617f1acaf0d12d8b177dbe6bf348
@anlunx Thanks in advance!
Copybara import of the project:
--
9dedb1ce2a620bae69c0fbaa8e5822ababfd52bc by Chao Che... | Chao | cchen104@amd.com | 1,695,043,942,000 |
4bc874a3ac339b16fd7735f1e6f8206f772ab6e1 | PR #5300: A new pass to optimize the AllGather->Binary_Op order sequence
Imported from GitHub PR https://github.com/openxla/xla/pull/5300
This is a new GPU SPMD optimization pass for the following pattern:
binary-op(all-gather(a), all-gather(b))
to
all-gather(binary-op(a, b))
Copybara import of the project:
--
77aa... | kushanam | kahmadian@nvidia.com | 1,695,047,980,000 |
191cf82bd66afbd32da0524b3bb506654eb57b23 | Integrate in the linaro fork of the TF docker | Michael Hudgins | michaelhudgins@google.com | 1,695,048,765,000 |
0a03fa6be53b915cdac744854c4f7ee59f364aa6 | [XLA:GPU] Triton GEMM rewriter: fix wrong check.
Fix forward for cl/566274725: IsEqualAt is templated and was not written correctly. The check was not related to the CL anyway.
PiperOrigin-RevId: 566298256 | Ilia Sergachev | sergachev@google.com | 1,695,048,020,000 |
ee70e8d292808c8264a349f5132ef1704e3ff0f6 | Update copyright headers | Michael Hudgins | michaelhudgins@google.com | 1,695,052,395,000 |
ba647c2fad0331a09b3924bf11e53b0141a53c86 | Resolved comment updated port.cc | mraunak | 83710963+mraunak@users.noreply.github.com | 1,695,055,661,000 |
0f6d8fd48c371d0e9debedbb1e4c574f9bf8b97b | Fix potential leak in SnapshotManager
In the error case, `snapshot_manager_` will not be deleted.
To solve that we can create the `unique_ptr` before calling `Start` instead of afterwards.
I just saw this by chance when looking into cl/566268125.
PiperOrigin-RevId: 566334445 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,056,412,000 |
57f33cabbe2dab00ac464ab20bf71d82c4c059fc | Rollback of PR #5300
PR #5300: A new pass to optimize the AllGather->Binary_Op order sequence
Imported from GitHub PR https://github.com/openxla/xla/pull/5300
This is a new GPU SPMD optimization pass for the following pattern:
binary-op(all-gather(a), all-gather(b))
to
all-gather(binary-op(a, b))
PiperOrigin-RevId: ... | Peter Hawkins | phawkins@google.com | 1,695,057,467,000 |
0e8c6fbea6b8999542f1969971ba439dfdce3f01 | Added placeholder for internal RPC option.
PiperOrigin-RevId: 566341265 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,057,693,000 |
dbd362e386ecb00d4b8c3b1e132ff78831ccb96c | Don't modify linked list while reading it.
PiperOrigin-RevId: 566341841 | Matthias Kramm | kramm@google.com | 1,695,057,792,000 |
b756908e266aca917bf698edb9da55d61e0b3c82 | Enable retries for initial connection by default for PSS to mitigate startup issues.
Add a test for startup fault tolerance.
Make the connection step more explicit; previously connection only happened within context initialization during a list_logical_devices call, which was unintuitive.
PiperOrigin-RevId: 56634426... | James Mullenbach | jmullenbach@google.com | 1,695,058,230,000 |
f85a264ba0e393d4863f89d3616d7884c99e0a0f | Fix race condition in SnapshotAssignmentManager
I stumbled upon this just by chance and found a solution.
That's how I understand it:
- Multiple instances of SnapshotManager share one instance of SnapshotAssignmentManager.
- SnapshotManager recently became thread-safe and multiple instances are being used from multi... | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,058,888,000 |
c581f9b1c392c1500eb03f56243c274bb380d347 | Rewire TensorFlow to rely on tf_keras target.
PiperOrigin-RevId: 566353493 | Francois Chollet | fchollet@google.com | 1,695,059,953,000 |
87694ad5b9f6867b35d2e1547a371e390f91176f | Fix AsyncCheckpoint issues.
1) Move `_copy_trackable_to_cpu` to `ShardedVariable` (from `ShardedVariableMixin`, which is inherited by other objects)
2) Fixed bug that excluded the handling of the TPUEmbedding object.
PiperOrigin-RevId: 566355521 | Katherine Wu | kathywu@google.com | 1,695,060,288,000 |
8618e98cb9ddd073908534fc4645a01f5328f60a | PR #5184: [XLA:GPU] fused attention runner clean up
Imported from GitHub PR https://github.com/openxla/xla/pull/5184
This is a follow up PR for cleaning up the redundant almost duplicated runners and MLIR op.
There are currently 4 runners and 3 MLIR op defined for fmha fwd. Different runner is chosen based on if ther... | Shanbin Ke | shanbinke@gmail.com | 1,695,060,952,000 |
49c00089c05337da5d9a0c537b1a93c989f990e1 | [XLA:GPU] Adds conditional return for SoftmaxRewriterTriton::Run to return false if no fusible diamond chains are matched.
PiperOrigin-RevId: 566368003 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,062,700,000 |
b3088931e988c9acf543126a9118b5d2e49c05c3 | Merge pull request #61894 from MichaelHudgins:arm-64-docker
PiperOrigin-RevId: 566369155 | TensorFlower Gardener | gardener@tensorflow.org | 1,695,063,637,000 |
5baed505a8d186c9c89dbb4bf5c10917d722d086 | Always consume the diagnostic handler status and return failure if it ever catches something the running of the passes does not.
PiperOrigin-RevId: 566370251 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,063,139,000 |
2177f8a43dc91d98e3e28e21f8e06cb10163e25b | Add a hlo-expand tool to run passes on HLOModule.
This tool lets you convert a HloModule from stdin or a file to another format,
run a set of expander passes, and dump the output to stdout or a file.
This expander tool is divided into the following steps:
1. Load HloModule from stdin or file.
2. Add a set of passes t... | Gunhyun Park | gunhyun@google.com | 1,695,063,975,000 |
57a86066fe57219d763d6783cb0f6e7d6490c4f5 | [XLA:GPU][NFC] Increase triton emitter test shard count.
PiperOrigin-RevId: 566375014 | Ilia Sergachev | sergachev@google.com | 1,695,064,076,000 |
59a577efd0e160b7ea3050784f7d37dcbb999e39 | [XLA] Disable async call syntax sugar when the async computation is nontrvial (i.e. does not expand to a single instruction) as that is not lossless. This is to ensure the two formats can roundtrip.
PiperOrigin-RevId: 566376729 | Ce Zheng | zce@google.com | 1,695,064,452,000 |
1c15b63a129bd782cc08b3e5985856a197e34641 | Make PeriodicFunction accept AnyInvocable. std::function doesn't support move-able callback.
PiperOrigin-RevId: 566385242 | Siqiao Wu | siqiaowu@google.com | 1,695,066,356,000 |
82b3111214216281b5aef214e53860703c6ccade | Move TSL/XLA headers to their original locations in tensorflow/include
PiperOrigin-RevId: 566385285 | Jake Harmon | jakeharmon@google.com | 1,695,066,366,000 |
716e1cc97751498963588f6059bc8f7ba342ab26 | Enable MLIR bridge in TF-TPU builds
PiperOrigin-RevId: 566387489 | Chandra Devarakonda | chandrasekhard@google.com | 1,695,066,797,000 |
110b707e5e13b58c44cabba2acb6a4977adad789 | Add an experimental option to `SavedModel` to suppress the addition of `SavedModel`-native save and restore ops to the `SavedModel`, for cases where users already build custom save/restore ops and checkpoint formats for the model being saved, and the creation of the `SavedModel`-native save/restore ops simply cause lon... | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,067,956,000 |
5d285bc149843d018281124fd4cb4c8efac520c2 | Annotate the dummy constant and Send/Recv pair created for control edges.
This change adds "/ctrl/" to the name of the dummy constant added by graph partitioning. As a result, it becomes easier to determine in a profile trace whether a Send/Recv edge was added due to a control or data dependency.
PiperOrigin-RevId: 5... | Derek Murray | mrry@google.com | 1,695,068,017,000 |
a7e764fc71a53157042890a4d30a05254257731d | Make diff_parser have visibility outside it's own package + fix formatting error
PiperOrigin-RevId: 566397309 | David Dunleavy | ddunleavy@google.com | 1,695,068,840,000 |
26c6758362088097ea5d8ab70fb8b7572b9e4c90 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/44239b5afd947ad56f13e65948254bd07ab2a384.
PiperOrigin-RevId: 566400358 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,069,484,000 |
3a67329732480f05fe02842b9051dab1a4083c50 | [XLA] Clarify code regarding dynamic literals
No functional change, let's just remove some ambiguity.
PiperOrigin-RevId: 566401107 | David Majnemer | majnemer@google.com | 1,695,069,626,000 |
3de44168950a5972ba4cfa7e3c6cbf4cffa67fe6 | Upgrade to LLVM 17, CUDA 12.2, and CuDNN 8.9.4
This is updating TF's default toolchain to LLVM 17, as well as
CUDA and cuDNN to the latest releases.
PiperOrigin-RevId: 566403707 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,070,211,000 |
28ab1668fe555262bc41b8377342b93c8dc5c1d5 | Don't register ops when TF APIs are not available
PiperOrigin-RevId: 566404976 | Haibo Huang | hhb@google.com | 1,695,070,474,000 |
2e7993a63a4c262e4a1e3453a9ffab1d0c151a4a | [XLA:GPU] Shard topk_kernel.cu.cc.
I've seen this file take over 5 minutes to build. Shard it by type.
PiperOrigin-RevId: 566409094 | Peter Hawkins | phawkins@google.com | 1,695,071,297,000 |
8818cf82858fc982e5dd8a58f92aaa1cd3abdffa | Integrate LLVM at llvm/llvm-project@14882d6b7440
Updates LLVM usage to match
[14882d6b7440](https://github.com/llvm/llvm-project/commit/14882d6b7440)
PiperOrigin-RevId: 566410108 | Fangrui Song | maskray@google.com | 1,695,071,507,000 |
f0d131703de11d51ee28925ebc4f0318fe08c44c | Delete `TransferPjRtBufferBetweenMemories` and replace it with `CopyToMemorySpace` which is more robust and fully async and transfers between any memory space.
PiperOrigin-RevId: 566420233 | Yash Katariya | yashkatariya@google.com | 1,695,073,699,000 |
94e73c964e3a834062d2db9fa44b35d97bd581dc | [tflite] Outline the stablehlo.scatter flatbuffer conversion logic
PiperOrigin-RevId: 566420937 | Majid Dadashi | majiddadashi@google.com | 1,695,073,858,000 |
00a17d7451a789a4df994dac7d616ce2f4438ff0 | Add error checking to make sure the API generator only writes to files that are passed in as output_files.
PiperOrigin-RevId: 566423310 | Brian Wieder | bwieder@google.com | 1,695,074,392,000 |
a86c6c5839ce95574663ee2e3bd8100ef07fbdf7 | Integrate LLVM at llvm/llvm-project@45735770ee87
Updates LLVM usage to match
[45735770ee87](https://github.com/llvm/llvm-project/commit/45735770ee87)
PiperOrigin-RevId: 566460975 | Fangrui Song | maskray@google.com | 1,695,083,806,000 |
1169ec8093ff91a02874cc81164809eb551b3957 | Remove :cudnn_wrappers BUILD target.
PiperOrigin-RevId: 566466412 | Peter Hawkins | phawkins@google.com | 1,695,085,488,000 |
dfcf1d40e46ce3c0a812a2703353da4f2d8e645d | Add third_party/triton/cl565664892.patch which is missing from a previous llvm integration
PiperOrigin-RevId: 566490078 | Fangrui Song | maskray@google.com | 1,695,093,754,000 |
4f638671ed7401e71ad7924bd329f8e0c6c6a861 | #tf.data Allow the `DataServiceClient` to increase `max_outstanding_requests` only when there is enough memory unless its value is specifically set by users.
PiperOrigin-RevId: 566497277 | Wilsin Gosti | wilsin@google.com | 1,695,096,173,000 |
bf9ebc7c3bc65062004b480f9a3f45b8de3050ec | Move TSL/XLA headers to their original locations in tensorflow/include
PiperOrigin-RevId: 566498953 | Jake Harmon | jakeharmon@google.com | 1,695,096,765,000 |
1af702b984eccc693352a5c3f437165a17981c06 | [XLA] Verify that async computations are trivial (contain only a root and parameter instructions).
PiperOrigin-RevId: 566509445 | Berkin Ilbeyi | berkin@google.com | 1,695,100,314,000 |
154eb9d524c5d9d5cb8a7d3ffe9b3b2b55bca475 | Internal Code Change
PiperOrigin-RevId: 566517153 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,102,503,000 |
298d4d5eeb402f6d7ea5be5a820d2fe9509e75ce | remove redundancy | Raunak | mayank.kumar.raunak@intel.com | 1,695,113,743,000 |
d26e5768abfd7f1708d6d07c818826d41cce7b93 | Merge branch 'mraunak/Clang_TF_Win' of https://github.com/Intel-tensorflow/tensorflow into mraunak/Clang_TF_Win | Raunak | mayank.kumar.raunak@intel.com | 1,695,114,050,000 |
c4952b207045131befabd30baca46ec686c698ef | Fixing bug introduced in cl/560954035. The `NVIDIA RTX A6000` is an sm_86 GPU, not sm_89.
PiperOrigin-RevId: 566557943 | Christian Sigg | csigg@google.com | 1,695,113,913,000 |
ef01cc1d01c7ff1a90a19608fa29b12bd3aa5ac1 | Update GraphDef version to 1624.
PiperOrigin-RevId: 566558824 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,114,121,000 |
fe2da355b5792fbfe3243fdcd6499b0e080f6cd7 | compat: Update forward compatibility horizon to 2023-09-19
PiperOrigin-RevId: 566559310 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,114,219,000 |
92cfa0e8d073f4bd5995fa0a9b9cff802950117a | Update XNNPack dependency
PiperOrigin-RevId: 566564208 | Zhi An Ng | zhin@google.com | 1,695,115,328,000 |
29abb6ac7121bed646f3cd832511372626e7916c | Update oneDNN to v3.2.1 for aarch64 and add fp32-bf16 JIT reorder patch | Fadi Arafeh | fadi.arafeh@arm.com | 1,694,783,450,000 |
5ead0b63566753d445c6766c0f365ed853396a9c | Integrate LLVM at llvm/llvm-project@4176ce61f156
Updates LLVM usage to match
[4176ce61f156](https://github.com/llvm/llvm-project/commit/4176ce61f156)
PiperOrigin-RevId: 566566342 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,115,915,000 |
6cbc40ce10da36d0519dd5a10b92c348cb3ea6f2 | [XLA:GPU][NFC] Make test TritonGemmTest.NondefaultOperandLayoutIsSupported faster.
And clean it up a little.
PiperOrigin-RevId: 566571071 | Ilia Sergachev | sergachev@google.com | 1,695,117,346,000 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.