hash stringlengths 40 40 | msg stringlengths 1 131k | author stringlengths 1 33 | email stringlengths 0 57 | date int64 1,447B 1,698B |
|---|---|---|---|---|
9a2a49232ca43b6254ecdce2f5549f1d71339b44 | Avoid deleting the boolean inputs from the top level FunctionDef's input signature when applying small_constants_optimizer.
PiperOrigin-RevId: 567661561 | Sagun Bajra | sagunb@google.com | 1,695,404,600,000 |
189556e2126cf36015cd33f2e3fa3f6a7abd0687 | Check if op kernel is XlaLocalLaunchOp before op kernel GetorCreate inside ExecuteOpConversion for MLRT
PiperOrigin-RevId: 567664326 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,405,177,000 |
85ee33e2395f0881c5f20be651ef3fc56a246347 | Replicate small constants so they don't need to be sent to their successors. A small constant is replicated to each of its successo...
PiperOrigin-RevId: 567666332 | Neel Kovelamudi | nkovela@google.com | 1,695,405,623,000 |
da49f51718a58759cd6d54d03bd7572aeef8c81c | PR #5822: [ROCm] fixes kernel name in rocm gpu executor
Imported from GitHub PR https://github.com/openxla/xla/pull/5822
ROCm fixes kernel name due to https://github.com/openxla/xla/commit/d04387fd4075af9d70858b2910cfa19a28a28fee
@akuegel @ezhulenev @ddunl Thanks in advance!
Copybara import of the project:
--
4cb21... | Chao | cchen104@amd.com | 1,695,406,189,000 |
34a5499d01293ee05c3c36d67aa2cc2891820489 | Add type hints for ops.Operation.
PiperOrigin-RevId: 567669383 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,406,288,000 |
efbc945baa1744389fba62202bf6dc6782914453 | [PJRT C API] Register "InspectSharding" custom partitioning handler in GPU plugin.
PiperOrigin-RevId: 567670200 | Jieying Luo | jieying@google.com | 1,695,406,450,000 |
3de450f2ebb7eca8b92ae24548ed11b53821c129 | [XLA:GPU] Fix accidentally broken Triton emitter test.
Addressing the test case just added in cl/567619889.
PiperOrigin-RevId: 567673155 | Ilia Sergachev | sergachev@google.com | 1,695,407,103,000 |
f60a692345e570d51b4dcb3578b475d229586933 | Add type hints under third_party/tensorflow.
PiperOrigin-RevId: 567675462 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,407,659,000 |
f2d3345f00add8504ef3297495c288e896b0212c | 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.
The previous vers... | Derek Murray | mrry@google.com | 1,695,407,884,000 |
cdfaf71501eda4df974e01f13b8af9dc335f03f6 | [xla:gpu] Set cublas workspace after calling cublasSetStream
We need to set workspace because new memory is allocated every time cublas calls is captured in cuda graph: https://docs.nvidia.com/cuda/cublas/index.html#cuda-graphs-support
We need to call it after set stream because cublasSetStream resets workspace:
http... | Anlun Xu | anlunx@google.com | 1,695,408,274,000 |
2ae6018db0ebfb4d1f8af0e6534a13296b3ecb5e | Internal change only
PiperOrigin-RevId: 567680498 | Gabriel Rasskin | grasskin@google.com | 1,695,408,856,000 |
75ebcea4fed412394bd99acd85d37c29fc9a1fc5 | Add found_fingerprint_on_load metric.
PiperOrigin-RevId: 567688573 | Adam Cogdell | adamcogdell@google.com | 1,695,410,802,000 |
11d6dacde62d48bdd9ba8af063ffbcb2df6ce0a7 | Integrate LLVM at llvm/llvm-project@6e3827af98fa
Updates LLVM usage to match
[6e3827af98fa](https://github.com/llvm/llvm-project/commit/6e3827af98fa)
PiperOrigin-RevId: 567691274 | Fangrui Song | maskray@google.com | 1,695,411,458,000 |
56a9f7f8404b95d62c1bae880fdaf88893166316 | Bincount with weights testing and documentation improvements
* Enable tests for Bincount with weights for `xla_gpu` and only skip for `gpu`.
* Update error message in GPU Bincount kernel to add `tf.function(jit_compile=True)` suggestion for using weights.
* Document equivalence between bincount and unsorted_segment ... | Edward Schwartz | schwartzedward@google.com | 1,695,411,655,000 |
5e2d5c50c2e9a7ecce00c8859d8f7fa1fddb752c | Internal Code Change
PiperOrigin-RevId: 567694546 | Gabriel Rasskin | grasskin@google.com | 1,695,412,261,000 |
7c8a95f2ab9b8996eccf5c33729018a45af467cb | Enable cross compilation for PJRT GPU compiler:
1. StreamExecutorGpuCompiler compiles w/o client.
2. Add StreamExecutorGpuExecutable (the unloaded pjrt executable).
3. Load StreamExecutorGpuExecutable to PjRtLoadedExecutable through `Load` API.
PiperOrigin-RevId: 567697879 | Shixin Li | shixinli@google.com | 1,695,413,126,000 |
2779ea10a44c93635e9f59f60cd320a293636440 | For embedding program key argument patching, allow ops with mini_batch_splits to
be outside of a tf_device.launch.
PiperOrigin-RevId: 567704317 | Matthias Kramm | kramm@google.com | 1,695,414,711,000 |
c273d635119a8b2082d6ace0bbcb0c5d614c84d2 | Fix PadOp lowering with negative padding amount
PiperOrigin-RevId: 567705787 | Logan Chien | loganchien@google.com | 1,695,415,084,000 |
af243fde7e770badda114281532a3cbd372becee | Remove unused `std::to_string()` in `GetMinibatchSplitsWithPhysicalReplicas` kernel.
PiperOrigin-RevId: 567709583 | Derek Murray | mrry@google.com | 1,695,416,030,000 |
4be8c8d2d20c7297fd3fc6a036a21098b830c9bc | Rewire TensorFlow OSS to use Keras 3 once the v3 release is live.
PiperOrigin-RevId: 567713807 | Francois Chollet | fchollet@google.com | 1,695,417,017,000 |
86c255ac098919ef6e72c957175a44f00d92e0c5 | PR #5830: [ROCm] Fix matching in launch_dimensions.hlo and slice_to_dynamic.hlo
Imported from GitHub PR https://github.com/openxla/xla/pull/5830
Copybara import of the project:
--
31c155a3110ac874912060f6cbc455b51f0c4858 by Dragan Mladjenovic <Dragan.Mladjenovic@amd.com>:
[ROCm] Fix matching in launch_dimensions.hl... | Dragan Mladjenovic | Dragan.Mladjenovic@amd.com | 1,695,417,145,000 |
4d076257f554c47c402549dec52ba2a10c8f85bc | This is an automatic update to an allowlist, work in progress.
PiperOrigin-RevId: 567714872 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,417,282,000 |
e2513c5f96a0504317960ad9a5df9dd381b278cb | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/2b29f4d9ec1bcbfaabc1d552c8e7aac9dc2809a1.
PiperOrigin-RevId: 567716100 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,417,602,000 |
11ead767cba2489056fc577fee350025b2fc768a | Fix build error on windows
PiperOrigin-RevId: 567716676 | Jake Harmon | jakeharmon@google.com | 1,695,417,756,000 |
5f7a3ed8d9672da3307865ec76102fead5296df8 | Set the default clang version to 17 or 16
PiperOrigin-RevId: 567721179 | Kanglan Tang | kanglan@google.com | 1,695,418,954,000 |
5f771ea1907d0974a2c78891de7c4f3fede8244c | [XLA] Initialize tuple shapes of async-done in dataflow analysis.
PiperOrigin-RevId: 567724401 | Berkin Ilbeyi | berkin@google.com | 1,695,419,785,000 |
44251d54e6c5f58e4909af0d948123d6634ec1ea | Integrate LLVM at llvm/llvm-project@46d5d264fc66
Updates LLVM usage to match
[46d5d264fc66](https://github.com/llvm/llvm-project/commit/46d5d264fc66)
PiperOrigin-RevId: 567728405 | Fangrui Song | maskray@google.com | 1,695,420,782,000 |
745a7fae7eafa9fb16849fd5ee9f24e0c2db3b7e | Fix TF-TPU build flags
PiperOrigin-RevId: 567730240 | Kanglan Tang | kanglan@google.com | 1,695,421,270,000 |
38c38ded948b16381b21b60d8cb9a53952bc8f03 | #tf-data-service Clean up log for distributed snapshot.
PiperOrigin-RevId: 567734740 | Yang Chen | yangchen@google.com | 1,695,422,449,000 |
ca20c978199289c147d0942861aac76b48cce33c | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/e8d8c19d9314439fb3e1c08936ba0dc6863d1ccc.
PiperOrigin-RevId: 567737020 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,423,125,000 |
d7027bd713999cdb3def5d365a0bca49b1b0f440 | Add a warning for Feature/table config names if not set. This field is required for running embedding lookup on latest tpu hardware.
PiperOrigin-RevId: 567738526 | Ziyin Huang | ziyinh@google.com | 1,695,423,551,000 |
8803725ae8ae13147e306b510e59caf2dbab58bd | [PJRT C API] Add a custom call C API to the priv of GPU PJRT_Api* and implement it in GPU plugin.
The priv of GPU PJRT_Api* has a chained of PJRT_Structure_Base which contains a type and a pointer to next PJRT_Structure_Base. Rename priv to extension_start.
PiperOrigin-RevId: 567740522 | Jieying Luo | jieying@google.com | 1,695,424,099,000 |
5c845a9e76a76474302bda77ec9c7f93579e8227 | Internal changes only
PiperOrigin-RevId: 567740639 | Kanglan Tang | kanglan@google.com | 1,695,424,133,000 |
8bffb7a2070abb941a12ea21d6f079a617cdaa58 | Fix a typo in `array_impl_test_lib`.
PiperOrigin-RevId: 567747222 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,426,147,000 |
5cd0a8fd0ef40b95726cb73ffb6484e4078a3a5e | Import ops/gradients_impl.py in eager/backprop.py to ensure that registered gradients are available for gradient lookup.
Remove nn_grad.py's dependency on eager/backprop.py to remove the cycle that would block this, by moving the nn gradients that depend on backprop to a new file.
PiperOrigin-RevId: 567749407 | Fiona Lang | flang@google.com | 1,695,426,825,000 |
f236ae3bca673d66d70e443dc8432a3e4e01bae2 | [stream_executor] NFC: Clean up xla/stream_executor:kernel build dependencies
PiperOrigin-RevId: 567763730 | Eugene Zhulenev | ezhulenev@google.com | 1,695,431,708,000 |
b0f08aa26133f3373d626381a58c04246dd35a11 | Add convenience method to get memory spaces by kind
PiperOrigin-RevId: 567781230 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,439,623,000 |
c2dfd70f7a3fa079dc53857571e4c406b254105c | [stream_executor] NFC: Clean up xla/stream_executor:kernel build dependencies
PiperOrigin-RevId: 567794680 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,445,853,000 |
9a4bde88bf5779ef0f39ef46eda4dc4e6090cf37 | Sub supports broadcasting up to 6 dimensions
PiperOrigin-RevId: 567812015 | Jared Junyoung Lim | jaredlim@google.com | 1,695,452,942,000 |
1b257eecced46265a7f3782c0cfaf70a32d73abd | Optimize BroadcastSub functions in TFLite reference kernel.
It removes redundant dimensions, compresses compressible dimensions, and handles broadcasting. Also vectorize int32 element-wise subtraction.
PiperOrigin-RevId: 567825531 | Jared Junyoung Lim | jaredlim@google.com | 1,695,459,469,000 |
c8fafddb7a53d779aa842ce075a2a0bea84711ee | compat: Update forward compatibility horizon to 2023-09-23
PiperOrigin-RevId: 567826030 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,459,718,000 |
35f4b9497cd316842d2d433b38be5ad28cb0e997 | Update GraphDef version to 1628.
PiperOrigin-RevId: 567826408 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,459,875,000 |
bb5c4abf1078ec2fde15f34052f7ba7d9251ab09 | [XLA] Fix build
A bad dep added in 198144efc4235162423939b2256505765f78e145 broke the build.
PiperOrigin-RevId: 567864100 | David Majnemer | majnemer@google.com | 1,695,480,698,000 |
51cc9dc4d9131e832bfc858dc59159e63ea8f116 | Adds back `allow_multiple_exports` which is a no-op to maintain `keras_export` compatibility with TF 2.13 and TF Nightly.
PiperOrigin-RevId: 567886699 | Ramesh Sampath | rameshsampath@google.com | 1,695,494,010,000 |
39447256a55bf5cd69907ae261eaa6ac98f80a1c | Internal Code Change
PiperOrigin-RevId: 567889304 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,495,710,000 |
e1538f18a95d9fed68a12a5aa2d46d1148588809 | Remove absl logging includes that are breaking build.
PiperOrigin-RevId: 567927507 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,520,083,000 |
359cf4552e205f0fb9cba966932273f1df3c8fd0 | Having live range dcheck respect threads to avoid false positive.
PiperOrigin-RevId: 567927880 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,520,324,000 |
c7eaebc0792ca5748bf85ec35c10f79851b78370 | compat: Update forward compatibility horizon to 2023-09-24
PiperOrigin-RevId: 567978384 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,546,146,000 |
fa4274041302f039f47527af1863e6574e73a42c | Update GraphDef version to 1629.
PiperOrigin-RevId: 567978540 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,546,195,000 |
bbb1f8e55b8c738b20a1d06b792c43157ac8c1cd | [xla:gpu] Set cublas workspace after calling cublasSetStream
We need to set workspace because new memory is allocated every time cublas calls is captured in cuda graph: https://docs.nvidia.com/cuda/cublas/index.html#cuda-graphs-support
We need to call it after set stream because cublasSetStream resets workspace:
http... | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,577,259,000 |
03813792494acc2512659e17773ec01969f9e633 | [XLA] Fix build for graph dumper for OSS builds
The stream_executor target does not bring in the symbol definitions for
PiperOrigin-RevId: 568101334 | David Majnemer | majnemer@google.com | 1,695,616,304,000 |
a984b89903194afde4cafab4658d1af8b5c5fc15 | Internal, build update for lexan.
PiperOrigin-RevId: 568111192 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,619,744,000 |
535288070845a6fb54e6d2a5b1b25549c80fedd2 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/549bf94d9643e0a1a9ea71949fde9e0a21dd30ea.
PiperOrigin-RevId: 568131952 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,626,393,000 |
014b270e98a2a0b045f7e9e8c7a917bf3aad898d | Corrected several typos
Documentation has been updated with correct words. Please have a look at this and do the needful Thank you! | sushreebarsa | 84765720+sushreebarsa@users.noreply.github.com | 1,695,626,946,000 |
c0c7efcff5bc6bffd2d0668f777196509d6971e8 | Update event_file_writer.py | sushreebarsa | 84765720+sushreebarsa@users.noreply.github.com | 1,695,627,408,000 |
99605b91f178a68d1eb8aa6bfb8bd9164a43d255 | Internal Code Change
PiperOrigin-RevId: 568147634 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,631,806,000 |
bf95ed963c863bfea33de7afad1dc47cafeac1d8 | Update summary.py | sushreebarsa | 84765720+sushreebarsa@users.noreply.github.com | 1,695,632,481,000 |
ac05bfd3f4d0733eeb4211f9cbee1d0313aba75a | Update GraphDef version to 1630.
PiperOrigin-RevId: 568149848 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,632,518,000 |
92213dbdb2a8c56b045aac745f466f9d15767435 | compat: Update forward compatibility horizon to 2023-09-25
PiperOrigin-RevId: 568149868 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,632,523,000 |
2d3445b9b9e01470094174ef83f2d7d700f02259 | Update run_models.py | sushreebarsa | 84765720+sushreebarsa@users.noreply.github.com | 1,695,633,439,000 |
27ea4421008f88d79220fe896ef75a0cb7f8b288 | Merge pull request #61883 from wamuir:strip-tsl-c-header
PiperOrigin-RevId: 568158672 | TensorFlower Gardener | gardener@tensorflow.org | 1,695,635,659,000 |
7ff47e53fb2b3d9567f362dcd70066add7c7721c | Integrate LLVM at llvm/llvm-project@f7bf99fb529f
Updates LLVM usage to match
[f7bf99fb529f](https://github.com/llvm/llvm-project/commit/f7bf99fb529f)
PiperOrigin-RevId: 568169974 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,638,776,000 |
9a0c0322dccc8a84a190a76ce9506eacd4d9c456 | Update TFRT dependency to use revision
http://github.com/tensorflow/runtime/commit/cf730a32058241ae24fd12fe0a9a0258faa0ca89.
PiperOrigin-RevId: 568177684 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,641,251,000 |
5edb674773776a44e24a73bcab94a38d7e15dab4 | Make a TF Dialect to Executor API, which is used in the Bridge Phase 1.
We'll separate out the current implementation in bridge.cc which currently always runs as part of the Bridge.
PiperOrigin-RevId: 568218119 | Mason Chang | masonchang@google.com | 1,695,653,577,000 |
b2a70ab4944fc4c54e8b6bf1b3bfcfc28d2abaad | [XLA:GPU] Allow fusing binary elementwise operations where exactly one operand is a splat constant into normalization diamonds.
PiperOrigin-RevId: 568238994 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,658,636,000 |
1ee265c2092b3dc631281965c3aa8967279a3e8e | Update the remote GCC toolchains to the newest container and defines to match the upgrade from CUDA 11.8 to 12.2
PiperOrigin-RevId: 568239285 | Michael Hudgins | michaelhudgins@google.com | 1,695,658,712,000 |
6655ca4631791a46df4ea7215aa0321200684216 | Definition for DFATAL is added to TensorFlow logging.
PiperOrigin-RevId: 568239531 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,658,769,000 |
9aaeb1bb04b12be62cad962f4b3adfcd344b535d | Keep dims when reducing gradients of broadcasted arguments to reduce size of dynamically-shaped graphs.
Motivation: Consider, for example, `z = x * y` where `x` has shape `[3, 1]` and `y` - `[3, 5]`. During backprop the upsteam gradient `gz` of `z` with shape `[3, 5]` is propagated as follows:
```
ix, iy = broadcast_... | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,659,414,000 |
ec9e1ee786feb60de67eecc4b4e2ae9c6dca493c | Add cuda 12.2 support to JAX
PiperOrigin-RevId: 568244230 | Yash Katariya | yashkatariya@google.com | 1,695,659,879,000 |
149cf817b927dc03c56ad57c4def9fb18fb5957a | Legalize some MHLO broadcasted binary element-wise ops (add, div, max, min, mul, pow, shift left, sub, atan2) to TF ops directly. tf.BroadcastTo op folder folds constants by default, this would increase the size of models converted from StableHLO because StableHLO requires explicit broadcasting. This change helps reduc... | Yishuang Pang | ypang@google.com | 1,695,661,611,000 |
036b68f244770a268ef3025bbeaac89793cf22bf | Add dependency from source node to VarHandleOp
This can guanrantee the graph's source node is connected to all source nodes in the graph.
PiperOrigin-RevId: 568252463 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,661,680,000 |
edf7215123c67d76199d099779137b974b6e1293 | [mlir][sparse][xla-cpu] Add rewriting rules to legalize sparse_tensor.convert
PiperOrigin-RevId: 568252598 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,661,702,000 |
8e7844a87b8bf640a3bf27040644060bce23119e | Add nccl to sigbuild and JAX Dockerfiles.
https://github.com/openxla/xla/pull/5850/files adds the ability to use a binary NCCL via a stub, just as with NVIDIA's other libraries. Make sure that NCCL is available in CI dockerfiles.
PiperOrigin-RevId: 568254770 | Peter Hawkins | phawkins@google.com | 1,695,662,118,000 |
26d09c4465b8d9376d574f7728092793a02deac9 | [PJRT C API] Implement framework side change for registering a custom call.
- Add a py extension to call the custom call C API.
- Change the implementation of register_custom_call_target to store handlers for the custom call targets and delays the registration until the handler for a xla platform is registered.
- Chan... | Jieying Luo | jieying@google.com | 1,695,664,272,000 |
397c205a3fa7b3c7c481f2224741c8b8a5e3523d | Enable imports such as `from tensorflow.keras import layers`
PiperOrigin-RevId: 568268550 | Francois Chollet | fchollet@google.com | 1,695,664,839,000 |
dc60587c40fcfa7e22111e76b2e4ec0a75d561a3 | [XLA:GPU] Support HLO input for AOT compilation
PiperOrigin-RevId: 568271995 | George Karpenkov | cheshire@google.com | 1,695,665,525,000 |
a914df7952a8ac3cce98266ded227a29a6315fa5 | [stream_executor] NFC: Clean up stream executor public libraries
Clean up and document stream executor "non-executor" public libraries.
https://github.com/openxla/xla/issues/5761
PiperOrigin-RevId: 568275550 | Eugene Zhulenev | ezhulenev@google.com | 1,695,666,213,000 |
4a3fd4ccb9c2f9e8f7b5dfc01fc40565578da3e2 | Merge pull request #61847 from joycebrum:master
PiperOrigin-RevId: 568281582 | TensorFlower Gardener | gardener@tensorflow.org | 1,695,667,779,000 |
d00f9eeb3105a7abe75ed7965b5a6b22e9aef596 | [stream_executor] Add back device_id_utils library
PiperOrigin-RevId: 568284525 | Eugene Zhulenev | ezhulenev@google.com | 1,695,668,103,000 |
3890704374c2a0ff95d4b3e2c684f19275f8071f | Move resource hoisting to before TPURewrite.
PiperOrigin-RevId: 568285914 | Mason Chang | masonchang@google.com | 1,695,668,396,000 |
9e29f611891202074bb669f2ffc5c865552644cd | Regenerate pyi files after mypy update
PiperOrigin-RevId: 568290052 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,669,241,000 |
e892ce92a745b2dc79349d6aa08eaf33e59e3744 | Support dynamic shape for zero-point-offset calculation for DotGeneral
PiperOrigin-RevId: 568291491 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,669,577,000 |
676af55749f95c486c6fce5b8b550c5412e28a5b | Use MLIR op lowering of FusedBatchNormV3 and FusedBatchNormGradV3 for tf2xla for
PiperOrigin-RevId: 568299087 | A. Unique TensorFlower | gardener@tensorflow.org | 1,695,671,389,000 |
2991aeb60931800b8103d6a89bc5d3b669f9ee73 | Use similarly signed types for comparison
In the downstream TFLM project, comparing int and size_t types results in sign-compare compiler warnings.
PiperOrigin-RevId: 568304134 | RJ Ascani | rjascani@google.com | 1,695,672,524,000 |
731a6beacd6ca2dfe3a24bbe47c18eb130307a3f | [TSL:CUDA] Improve the "I didn't find this symbol" stub for cudaGetErrorString() and cudaGetErrorName().
These functions return strings; return a generic error here instead of returning an invalid integer value.
Fixes crash in JAX CI.
PiperOrigin-RevId: 568308538 | Peter Hawkins | phawkins@google.com | 1,695,673,499,000 |
257a679f7c80046ce6398413d2db767b3e15aa08 | [xla:gpu] xfeed: add missing absl/synchronization dep
PiperOrigin-RevId: 568310784 | Emilio Cota | ecg@google.com | 1,695,673,991,000 |
f1b7a6a604dc847ddf6edb90f25c8a42f2434c58 | Legalize EmptyTensorList as custom TensorListReserve with length 0.
PiperOrigin-RevId: 568311796 | Luke Boyer | lukeboyer@google.com | 1,695,674,231,000 |
a1543a3a49eedb424967f33ba7b876423df1db60 | fix for Ubuntu CPU test cases | Raunak | mayank.kumar.raunak@intel.com | 1,695,675,208,000 |
3bc2b252352279036c5957b44156d9366fda0474 | [stream_executor] Add GPU graph handle to GpuCommandBuffer
https://github.com/openxla/xla/issues/5857
PiperOrigin-RevId: 568312557 | Eugene Zhulenev | ezhulenev@google.com | 1,695,674,407,000 |
bbda501caf7e83e7d390c372647ec23864a9c65c | Add new context type for serving
PiperOrigin-RevId: 568324654 | Feng Wang | wffw@google.com | 1,695,677,186,000 |
f21782ea1d8b5bb25ce2a5146950677aff650160 | [tsl] Add the option to obtain NCCL via a stub, rather than linking it statically or dynamically.
NCCL via a stub is enabled if the environment variable TF_NCCL_USE_STUB=1 is set.
The intent is to use this option in JAX to reduce the size of the CUDA wheels. NCCL takes up about 80MB in each compiled JAX wheel and tak... | Peter Hawkins | phawkins@google.com | 1,695,682,002,000 |
8db216ce8c4043ab6a767010d313ff896b5bd8fb | fixed tests failure on Ubuntu CPU | Raunak | mayank.kumar.raunak@intel.com | 1,695,683,437,000 |
450dec35448a73b3fcb5d4f82108d5fdcb3f59b4 | Internal change, add some checks on the sparseTensor format checking.
PiperOrigin-RevId: 568349775 | Ziyin Huang | ziyinh@google.com | 1,695,683,187,000 |
9ad61cacb80b689addff053fe8181d59c2fbdc8b | format fix | Raunak | mayank.kumar.raunak@intel.com | 1,695,683,840,000 |
0bd205eeb7db15238aa795243cc07e401e97e71e | Split long-running test cases in mul_test into subshards.
This improves parallelism by evening out the sizes of the
different shards and thus speeds up this sharded test.
PiperOrigin-RevId: 568351990 | Jared Junyoung Lim | jaredlim@google.com | 1,695,683,691,000 |
9668ea0809b56288400d9dadaae57678043046b3 | Legalize some MHLO broadcasted select op to TF SelectV2 op directly. tf.BroadcastTo op folder folds constants by default, this would increase the size of models converted from StableHLO because StableHLO requires explicit broadcasting. This change helps reduce model size by removing unnecessary broadcasts at legalizati... | Yishuang Pang | ypang@google.com | 1,695,683,829,000 |
55b7276ba525f322d7bd7773f61f0938a13292d1 | #tf-data-service Scale up `"no_compression"` experiment to 50%.
PiperOrigin-RevId: 568353683 | Matt Callanan | mpcallanan@google.com | 1,695,684,103,000 |
882e787f3d989df283811fd8309177601eccd6e9 | [PJRT C API] Split callback and py_client_gpu out from py_client.
This is to prepare separating GPU into a plugin. The plugin will use py_client_gpu to register custom callback for `xla_python_gpu_callback`.
PiperOrigin-RevId: 568368076 | Jieying Luo | jieying@google.com | 1,695,687,868,000 |
c475f59d1a3c735a94ed8078983fc0119259bef1 | Create AotCompileSavedModel function to produce AotResult w/o writing to disk.
PiperOrigin-RevId: 568379208 | Shixin Li | shixinli@google.com | 1,695,691,086,000 |
b53353deeecec5c99689c8d63a9e719137612f6a | Update the containers which contains a rebuilt image which removes bazel from jax's docker image and use the ones preinstalled in the sigbuild image
PiperOrigin-RevId: 568381631 | Yash Katariya | yashkatariya@google.com | 1,695,692,035,000 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.