Changes

#692 (May 22, 2020, 11:47:00 AM)

1. Enabe BF16 SoftmaxGrad(Sum), and fix accuracy by accum type. — teng.lu / detail
2. Fixed eager mode gradient checkpointing by eliminating unecessary persistence of intermediate activations in memory — paiajay / detail
3. Removed API change and added tests. — paiajay / detail
4. Fix issue in tf.image.extract_glimpse — yong.tang.github / detail
5. Add test cases for tf.image.extract_glimpse — yong.tang.github / detail
6. Fix test failure caused by API changes in tests — yong.tang.github / detail
7. Fix test failure due to changes of the fix for centered=False and normalized=False — yong.tang.github / detail
8. Fix incorrect doc test — yong.tang.github / detail
9. Fix incorrect doc example with centered=False — yong.tang.github / detail
10. Use ExtractGlimpseV2 and ExtractGlimpse to make sure C++ kernel is backward compatible — yong.tang.github / detail
11. Expand test case to cover both old kernel (ExtractGlimpse) and new kernel (ExtractGlimpseV2) — yong.tang.github / detail
12. Add api_def_ExtractGlimpseV2.pbtxt — yong.tang.github / detail
13. Update API golden — yong.tang.github / detail
14. Reroute tf.compat.v1.extract_glimpse to use gen_image_ops.extract_glimpse (old API) — yong.tang.github / detail
15. Update RELEADE.md to capture the breaking change of `tf.image.extract_glimpse` — yong.tang.github / detail
16. Update doc example of v1 to keep old behavior with usage of tf.compat.v1.image.extract_glimpse — yong.tang.github / detail
17. Modified recompute_grad to handle fwd mode diff — paiajay / detail
18. Addressed PR comments — paiajay / detail
19. Addressed PR comments — paiajay / detail
20. Reorganized tests for recompute grad — paiajay / detail
21. Support BF16 Softmax and add UT. — teng.lu / detail
22. Fixed tests — paiajay / detail
23. Added grad_wrapper to accomodate graph mode — paiajay / detail
24. Fixed test — paiajay / detail
25. Addressed PR comments — paiajay / detail
26. Fixed sanity test issue — paiajay / detail
27. Fixed test — paiajay / detail
28. Allow rounding loss in Reduce UT. — teng.lu / detail
29. Fixed test — paiajay / detail
30. Fixed test — paiajay / detail
31. Update TF Lite roadmap — gardener / detail
32. [TF:TRT] Delay the computation of GraphProperties used by ConvertAfterShapes. — gardener / detail
33. Add tf.Yield and tf.IfRegion op to model non-functional conditionals — gardener / detail
34. Prevent extra downloads for the xtensa-xpg target. — gardener / detail
35. Limit FillOp custom folder to int and float types to avoid crash — gardener / detail
36. Internal change — gardener / detail
37. Remove illegal BroadcastTo op compiler tests — gardener / detail
38. [TF] Add support for more than one outer batch dimension to tf.nn.convolution. — gardener / detail
39. Enable skipped test as Variable.assign(name=xxx) now works with CentralStorage — gardener / detail
40. Support ShardedVariable in `tf.keras.layers.Embedding`. — gardener / detail
41. Fix makefile benchmark script to build on mac os. — gardener / detail
42. Move the feature_column.LinearModel to estimator which is the only caller for it. — gardener / detail
43. cupti tracer allow device synchronization before gputracer stop. — gardener / detail
44. [tf.data service] Perform tf.data service compression within tf.data. — gardener / detail
45. Implement a pass that converts readonly reference variables to the corresponding — gardener / detail
46. Fix for hello_world memory error — gardener / detail
47. Switches keras.backend.placeholder + keras.backend.function to build a keras model when running eagerly (instead of trying to directly lift ops out of a graph into a concretefunction). Allows us to strip most of EagerDefinedExecutionFunction from the keras backend. — gardener / detail
48. Internal change — gardener / detail
49. Take device locality into account during prioritization. — gardener / detail
50. Nit: Fix some typos. — gardener / detail
51. [NFC] Fix typos and adopt Google style variable names — gardener / detail
52. Enable gradient tests for tf.linalg.cholesky in eager mode. — gardener / detail
53. Exporting CategoryEncoding layer. — gardener / detail
54. Support the pybind11 `_dtypes.DType` in `tf.as_dtype`. — gardener / detail
55. [TF] Add support for more than one outer batch dimension to tf.nn.convolution. — gardener / detail
56. [tf.data service] Apply dataset options to tf.data service side datasets. — gardener / detail
57. Disable two tests on windows due to image issue. — gardener / detail
58. Disable flaky test. — gardener / detail
59. switch capture_tpu_profile to new api of profiler_client. because some options is dropped, therefore two flags are deprecated. — gardener / detail
60. Add memory cost breakdown per operation type and memory space to OpMetrics. — gardener / detail
61. PR #39548: [INTEL MKL] Fix conv_ops_test and remapper_test — gardener / detail
62. Adding uint32 support for more variable related operations. — gardener / detail
63. Update ops-related pbtxt files. — gardener / detail
64. Reapplying #39042 to avoid breaking some internal users. — gardener / detail
65. Fix SideEffecting semantics of xla_hlo ops without regions. — gardener / detail
66. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
67. Fix infinite loop in GetMatchingPaths in GCS — gardener / detail
68. Handle case when input is complex64 type. — gardener / detail
69. Simplify and address missing features in TPU Extract Head Outside Compilation pass. — gardener / detail
70. Enable 1st order gradient tests for tf.linalg.svd in eager mode. — gardener / detail
71. Enable tests for tf.linalg.matrix_logarithm in eager mode. — gardener / detail
72. Don't disable all TPU tests just the ones that fail. — gardener / detail
73. Fix issues where index_lookup was improperly handling hard vocab caps. Add tests. — gardener / detail
74. Add bfloat16 support for SparseSegmentMean*/SparseSegmentSqrtN* — gardener / detail
75. Improve diagnostic when a mutable global tensor is found — gardener / detail
76. Reduce Layer.__call__ overhead by ~5%. — gardener / detail
77. Test hello world example binary can run — gardener / detail
78. Temporarily blacklist keras model_coverage_lib_tests from TF2 testing — gardener / detail
79. Add vlogging for compression data size. — gardener / detail
80. Test that micro speech example binary can run — gardener / detail
81. FloorDiv and FloorMod dynamic shape support — gardener / detail
82. Improve side-effecting semantics of xla_hlo ops with regions. — gardener / detail
83. Pfor: support vectorizing tensorlists captured by while_loop. — gardener / detail
84. Make tf_saved_model-related passes check their input invariants. — gardener / detail
85. Update ops-related pbtxt files. — gardener / detail
86. Integrate LLVM at https://github.com/llvm/llvm-project/commit/1108f5c737db — gardener / detail
87. Remove superfluous tracemes in compression_utils. — gardener / detail
88. Bump open source llvm revision to 1108f5c737dbdab0277874a7e5b237491839c43a — gardener / detail
89. Internal TF change. — gardener / detail
90. [Grappler] Do not add data inputs after control inputs in ImplementationSelector — gardener / detail
91. Add bfloat16 support for SparseSegmentMean*/SparseSegmentSqrtN* — gardener / detail
92. Internal change — gardener / detail
93. [XLA] Fixup the bug in tautological compare simplifier, as spotted by Sanjoy — gardener / detail
94. [TF/XLA] Support F64 conversion for tf.cumsum — gardener / detail
95. Fix TFLiteConverter2 API Documentation to read frozen_graphs. — gardener / detail
96. [tfdbg2] Document the need to call set_soft_device_placement(True) on TPUs — gardener / detail
97. Add tf.function test for device placement logging — gardener / detail
98. Update sparse input documentation. — gardener / detail
99. Rearrange the binary __operator__ code in TensorFlow to be more dispatch-friendly. — gardener / detail
100. Disable multi-threaded Conv optimizations w/ non-const filters — gardener / detail
101. [XLA] Introduce asynchronous collective-permute (CollectivePermuteStart and CollectivePermuteDone) HLO opcodes. — gardener / detail
102. Disable multi-threaded Conv optimizations w/ non-const filters — gardener / detail
103. [XLA] Add AllGather to the HLO matchers. — gardener / detail
104. [XLA] CopyStart/CopyDone times should use exclusive indices, not inclusive. — gardener / detail
105. Reduce Layer.__call__ overhead by ~5-10%. — gardener / detail
106. [XLA] Use all-gather in SPMD to replicate a tiled tensor. — gardener / detail
107. compat: Update forward compatibility horizon to 2020-05-22 — gardener / detail
108. Update GraphDef version to 409. — gardener / detail

#691 (May 21, 2020, 11:47:00 AM)

1. docs: add tip to prefer tf.shape(x) over x.shape when writing custom layers/models — github / detail
2. Fix typo preventing compilation — felix.klee / detail
3. shorten tf.shape docstring — github / detail
4. [XLA:CPU] Plumb through a minimal emitter for matmuls using the mlir linalg dialect — gardener / detail
5. Use tf.io instead of os for checking if paths exist in Keras applications. — gardener / detail
6. [XLA:CPU] Plumb through a minimal emitter for matmuls using the mlir linalg dialect — gardener / detail
7. Update StopGradients's overly strict verification — gardener / detail
8. Use `--ignore-installed` instead of `--force-reinstall` when reinstalling tf-estimator-nightly. — gardener / detail
9. Add xla_hlo.case op for  indexed conditional HLO. — gardener / detail
10. When calling connect_to_cluser, if the options are identical and there is no renaming of local device, reuse existing local DeviceManager, otherwise we keep the old DeviceManager around to allow the old Tensor created to be usable. — gardener / detail
11. Nit: Remove trailing whitespace. — gardener / detail
12. [TF] Add support for more than one outer batch dimension to tf.nn.conv2d. — gardener / detail
13. Fix sha256sum issue of mkl_dnn/oneDNN — yong.tang.github / detail
14. Fix missing ',' in workspace.bzl — yong.tang.github / detail
15. Auditing some const ref params for the xtensa kernels. — gardener / detail
16. Handle TPU inputs to OutsideCompiled parallel_execute regions. — gardener / detail
17. [XLA] Instead of limiting outstanding async copies, limit outstanding prefetch and evictions — gardener / detail
18. Implement the llvm lowering for the customcall sliceToDynamic and padToStatic for multi-dimensional array on XLA:CPU — gardener / detail
19. [TF MLIR SI] Don't constant fold, only consider result of folding — gardener / detail
20. SyncOnReadVariable.assign() should return Tensor — gardener / detail
21. Replace absl::StrCat with llvm::formatv when creating communication key in TPUExtractOutsideCompilation pass (NFC). — gardener / detail
22. Prune redundant control inputs early in model_pruner, since they may prevent deletion of trivial nodes. — gardener / detail
23. Remove TODO and insert helpful comment to ResizeInputTensor() — gardener / detail
24. [XLA] Fix async copy statistics to include non-entry computations. — gardener / detail
25. Provide a more informative error message when the bazel version check fails — gardener / detail
26. Move the _BaseFeatureLayer back to Keras. — gardener / detail
27. Replace _TPUCompileMlir placeholder ops with correct compile op in parallel_execute regions. — gardener / detail
28. Clean up unneeded lines from  BUILD file. — gardener / detail
29. Correct mistaken ms to ns conversion change. — gardener / detail
30. Narrow down argmin/argmax contract to always return the smallest index for ties — gardener / detail
31. Removing redudancy in device mapping logging — gardener / detail
32. Hexagon Delegate: — gardener / detail
33. Fix BufferAndPlanClearingTest — gardener / detail
34. Handle complex data types in tf.SigmoidOp and tf.SigmoidGradOp lowering — gardener / detail
35. [TF/XLA] Implement TensorListConcat kernel conversion for XLA — gardener / detail
36. Update Eigen to: https://gitlab.com/libeigen/eigen/-/commit/cc86a31e20b48b0f03d714b4d1b1f50d52848d36 — gardener / detail
37. Clean up saving tests for TextVectorization and add the layer to default serialization. — gardener / detail
38. Internal visibility rule change. — gardener / detail
39. Fix the formatting of TensorArray. __doc__ indents anything after the first line doc. Dedent the rest of the doc after the first line so that it is formatted properly on the site. — gardener / detail
40. Align example models to 64-bit boundaries to guarantee correctness for all 64-bit flatbuffer accesses.  Aligning 64-bit datatypes to 32-bits can cause memory errors on some architectures. — gardener / detail
41. Convert TF_SavedModel to a direct pointer to tensorflow::SavedModelAPI. This saves us an extra allocation when loading a savedmodel, and extra indirection on all saved model functions. — gardener / detail
42. use bincount_v2 in tf.math.bincount, support axis and binary_output, and support int64. — gardener / detail
43. Add doctests to lookup layers. — gardener / detail
44. DataFormatVecPermute accepts a vector of size 2. — gardener / detail
45. Run build_cleaner to fix dependencies. — gardener / detail
46. Enable gradient tests for tf.linalg.qr in eager mode. — gardener / detail
47. Moving TPU ops components as TPU kernels library. — gardener / detail
48. Export Hashing layer. Add `separator` for CategoryCrossing and tf.sparse.cross. Add benchmarks for hashing. — gardener / detail
49. Add `inference_input_type` and `inference_output_type` flags in TF 2.x TFLiteConverter (backward compatible with TF 1.x) to support integer (tf.int8, tf.uint8) input and output types in post training full integer quantized models. — gardener / detail
50. Update optimize pass to adopt the shape constraints of binary ops — gardener / detail
51. Have sanity build output a Bazel test summary XML file. — gardener / detail
52. Add `inference_input_type` and `inference_output_type` flags in TF 2.x TFLiteConverter (backward compatible with TF 1.x) to support integer (tf.int8, tf.uint8) input and output types in post training full integer quantized models. — gardener / detail
53. Add op sanity checks to the following TFLite ops: — gardener / detail
54. pfor: Add support to vectorize TensorList operations: — gardener / detail
55. Remove TFLite Java runtime dependency in the metadata java lib — gardener / detail
56. Adds necessary hooks to load a TPU-specific shared library. — gardener / detail
57. Add dataset element compression ops. — gardener / detail
58. Update person_detection_experimental model — gardener / detail
59. Fuse tf.text WhitespaceTokenizer to tflite custom op — gardener / detail
60. Update ops-related pbtxt files. — gardener / detail
61. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
62. Remove some unused methods. — gardener / detail
63. Take proto by value. — gardener / detail
64. Update GraphDef version to 408. — gardener / detail
65. compat: Update forward compatibility horizon to 2020-05-21 — gardener / detail
66. Export CategoryEncoding keras preprocessing layer. — gardener / detail
67. Add Int8 PRelu operation to TFLM. — gardener / detail
68. Internal change — gardener / detail

#690 (May 20, 2020, 11:47:00 AM)

1. Fix a bug related to build TF Lite on RPI Zero. — github / detail
2. Cross and native compilation of TFLite for RPI — marcin.sielski / detail
3. Address issues identified during review — marcin.sielski / detail
4. Add clone step to native build instrunction. — marcin.sielski / detail
5. Generate a cubin header for tanh. — gardener / detail
6. Simplify cuda toolchain config — gardener / detail
7. Fix bug where dispatch broke for ops that define an argument named 'op'. — gardener / detail
8. [XLA] Improve cost analysis for while loops. — gardener / detail
9. Fix ops pbtxt — gardener / detail
10. [TF saved_model_cli] Allow user to set target_cpu for xla aot compilation. — gardener / detail
11. Fix condition for activation function. FullyConnected in hexagon delegate doesn't support activation currently. — gardener / detail
12. Update TextVectorization to use internal layer adapt calls instead of its own combiner. — gardener / detail
13. Internal visibility change — gardener / detail
14. Note the reason why tests in numerics_test.py are marked v1-only — gardener / detail
15. Add a Compare() builder that is compatible with omitting broadcast_dimensions with the same ordering as the other binary ops. — gardener / detail
16. Integrate LLVM at https://github.com/llvm/llvm-project/commit/2e499eee5884 — gardener / detail
17. Minor fix to include order. — gardener / detail
18. Add TPUEmbedding mid level API. — gardener / detail
19. Simplify nccl_configure. — gardener / detail
20. Rename xla_hlo.conditional to xla_hlo.if. — gardener / detail
21. Delete Python ScopedAnnotation — gardener / detail
22. Remove implicit broadcasting from xla_hlo binary elementwise ops. — gardener / detail
23. Do TraceMe kwargs encoding in C++ — gardener / detail
24. Add an 'invert' arg to lookup layers. — gardener / detail
25. Remove unused experimental APIs — gardener / detail
26. Update the comment for the normalization parameters — gardener / detail
27. Add bot comment for cuda and windows related build and install issues. — gardener / detail
28. Disable //tensorflow/python/eager:function_test on macos for now. — gardener / detail
29. Enable TensorCore embeddings for training via FeatureColumnV2. — gardener / detail
30. Fix performance regression involving trainable check in batchnorm. — gardener / detail
31. Test ConvertActivation in dynamic shape mode — tfeher / detail
32. Fix pylint errors — gardener / detail
33. Use the nonlocal mechanism for if statements. This is the same mechanism used by for and while loops and it allows reusing much of the code. — gardener / detail
34. Don't try t gather from empty tensors — gardener / detail
35. Canonicalize tf.Select to tf.SelectV2. — gardener / detail
36. Hexagon delegate: — gardener / detail
37. Fix parameter check for batchdim in Gather. — gardener / detail
38. Improve the bucket generation in Boosted Trees to avoid returning more than requested buckets. — gardener / detail
39. Split delegate-specific interpreter tests into a separate file — gardener / detail
40. adjust gather ops launch config. for NCF model, this means ~20% gain. (due to grid size from 80->160 on volta). — gardener / detail
41. Add a folder for numpy API — gardener / detail
42. Allow tile op to work on variant dtype. — gardener / detail
43. pfor: Enable handling VariableShape — gardener / detail
44. Use enum instead of string to specify the context type. — gardener / detail
45. Fix nightly breakage (macos, linux). — gardener / detail
46. [Profiler] Add optimization advice for reducing the host-to-TPU data transfer time. — gardener / detail
47. Add a global resource manager for TPU specific operations. — gardener / detail
48. [XLA] Fix some all-gather issues. — gardener / detail
49. Internal change — gardener / detail
50. Add basic support for top level function/NameAttrList attributes in Graph -> TF MLIR importer. — gardener / detail
51. Use functools partial rather than lambda when calling initializer. Patch TPU Embedding API to prioritize extracting shape from functools.partial objects. — gardener / detail
52. Deduplicate Python TraceMe implementations — gardener / detail
53. Include missing dependency. — gardener / detail
54. Automatically cast int32 inputs to int64 if the table key expects int64. — gardener / detail
55. Rename numpy to numpy_ops — gardener / detail
56. Look up eager client directly from target in eager cluster FLR. — gardener / detail
57. Disable flaky tensorflow/c/eager:c_api_remote_test on asan — gardener / detail
58. Adding benchmarks for image preprocessing layers. — gardener / detail
59. Update TextVectorization docstring. — gardener / detail
60. Set static shape information for reshape layer. — gardener / detail
61. [XLA] Strength reduce cvt(pred) / bcast(f32) to bcast(1 / f32) * cvt(pred) — gardener / detail
62. Fix formatting of file — gardener / detail
63. Disable flaky TFRT tests. — gardener / detail
64. Build pip_package with Bazel — gardener / detail
65. [XLA:PARSER] Fix parser to read U64 types properly — gardener / detail
66. Add shape inference pass before legalization: prepare-tf may introduce ops those do not have shaped tensor yet, we need to insert a shape inference pass to make proper shape propagation. — gardener / detail
67. [XLA] Unbreak the OSS build — gardener / detail
68. Update GraphDef version to 407. — gardener / detail
69. compat: Update forward compatibility horizon to 2020-05-20 — gardener / detail
70. [XLA:CPU] Plumb through a minimal emitter for matmuls using the mlir linalg dialect — gardener / detail

#689 (May 19, 2020, 11:47:00 AM)

1. Use provided host name/ip instead of localhost if possible — sunfei.china / detail
2. Edit according to PR comments — sunfei.china / detail
3. ROCm 3.5 (hip-clang) build fixes — eugene.kuznetsov / detail
4. fix conv_ops_test and remapper_test — guozhong.zhuang / detail
5. Test autograph indirect tf.map_fn decorator — bhack / detail
6. Fix missing return — bhack / detail
7. Fix missing self — bhack / detail
8. Add test and remove decorator — bhack / detail
9. restore a remove export — bhack / detail
10. Update SECURITY.md — github / detail
11. Verifiy differences with test annotation — bhack / detail
12. Let test to fail — bhack / detail
13. Test autograph transform of fn — github / detail
14. Remove original fn call — github / detail
15. Fix lint and improve readibility — bhack / detail
16. Reviewer requests — eugene.kuznetsov / detail
17. numerics_test.py: Move tfdbg2-specific test methods to debug_v2_ops_test.py — gardener / detail
18. [tfdbg] Fix source_utils_test in Python 3.8+ — gardener / detail
19. Optimize multiply by quantize multiplier. — gardener / detail
20. Resolve trivial aliases for portable TensorFlow targets. — gardener / detail
21. Integrate LLVM at https://github.com/llvm/llvm-project/commit/a2a4e5aae894 — gardener / detail
22. [TF:TRT] Enhance InstantiateBuildAndRun to support the case where the input — gardener / detail
23. Optimize quantized mul. — gardener / detail
24. [TF:TRT] Add utilities for converting between TF types and TRT types. — gardener / detail
25. Internal change — gardener / detail
26. [TF:TRT] Implement cast from fp16 to fp32 with IIdentityLayer. — gardener / detail
27. Update release notes for the 1.15.3, 2.0.2 and 2.1.1 patch releases. — gardener / detail
28. Tweak round_to_bfloat16 to make it vectorizable. — gardener / detail
29. Report remote target name for worker service RPCs. — gardener / detail
30. Add legalization from hlo.dot to lhlo.dot — gardener / detail
31. Enable (non-gradient) tests of tf.linalg.cholesky in eager mode. — gardener / detail
32. Support int8 in tflite_convert — gardener / detail
33. Ensure that tf_py_test tfrt test is not enabled for open source build by introducing tfrt_enabled_internal flag. — gardener / detail
34. Document new methods to enable XNNPACK engine in TFLite — gardener / detail
35. enable device tracer test. — gardener / detail
36. [XLA:Python] Delete deprecated methods from XLA:Python API. — gardener / detail
37. Hexagon Delegate — gardener / detail
38. [XLA] Fix rendering of the RngBitGenerator description table — gardener / detail
39. Enable tests for tf.linalg.lu in eager mode. — gardener / detail
40. Enable tests for tf.linalg.matrix_square_root in eager mode. — gardener / detail
41. Rename TraceMe::SetMetadata to TraceMe::AppendMetadata and add lambda overload. — gardener / detail
42. [XLA] Simplify tautological compares (and (< x A) (< x B)) to (< x A) when `a <= B` holds. — gardener / detail
43. [tf.lite] Use in-process conversion when the new converter is used — gardener / detail
44. Edit Hexagon documentation to reflect new supported models — gardener / detail
45. Enable tests for tf.linalg.tensordot in eager mode. — gardener / detail
46. Return a meaningful error for dynamic shape inputs with outside compilation head extraction in TPUs. — gardener / detail
47. NFC: — gardener / detail
48. Rename `_get_closest` to more accurately reflect what it does. — gardener / detail
49. Modify signature of layout_config(). — gardener / detail
50. In TF-TFRT integration, C API will get dtype from underlying fallback tensor directly if the tfrt dtype is Unsupported. This is used to support dtypes that are not natively implemented in TFRT (e.g. DT_RESOURCE). — gardener / detail
51. Support running a tf.function with packed variable inputs both locally and remotely. — gardener / detail
52. Updates GPU delegate documentation with experimental quant support — gardener / detail
53. Re-enable signal kernel tests on py38 — gardener / detail
54. Allow static result shape for unranked operand in shape verifier — gardener / detail
55. Fix argument check tests to work in eager mode — gardener / detail
56. Add lowering from xla_hlo/lhlo reverse op to Linalg. — gardener / detail
57. [XLA] Fix alternate memory allocation of conditional operands. — gardener / detail
58. Rename TransformTensorV2 op to TransformTensorBilinearV2 op. — gardener / detail
59. Add SparseCrossV2 which supports strong_hash with salt, and fingerprint doens't — gardener / detail
60. Remove the unnecessary address-returning operator and lamda expression. — gardener / detail
61. Hexagon Delegate — gardener / detail
62. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
63. Disable flaky tensorflow/c/eager:c_api_test — gardener / detail
64. Move compression_utils to core/data. — gardener / detail
65. Adding skip record functionality to snapshot utils. — gardener / detail
66. Move tf.keras.layers.featureDenseFeature back to Keras package. — gardener / detail
67. Fix TF_ConcatV2Op conversion pattern when the axis is a I64 Tensor. — gardener / detail
68. NFC: — gardener / detail
69. Make serialization of node_def.attr() deterministic. — gardener / detail
70. [tf.data] Remove several unnecessary lines in the test. — gardener / detail
71. Introduce TraceMeProducer and TraceMeConsumer. — gardener / detail
72. slightly improve quantized max performance. — gardener / detail
73. [TF] Add eager microbenchmark for conv2d. — gardener / detail
74. Remove trivial quantize op — gardener / detail
75. Fix bug in Canonicalizer folder function for ArithmeticOp. — gardener / detail
76. Enable more TFRT tests. — gardener / detail
77. Reorder functions in an effort to group utility functions that use symbols defined in values.py and are used by classes defined in values.py. — gardener / detail
78. Integrate LLVM at https://github.com/llvm/llvm-project/commit/a6be4d17e349 — gardener / detail
79. compat: Update forward compatibility horizon to 2020-05-19 — gardener / detail
80. Update GraphDef version to 406. — gardener / detail
81. Move GraphWithDequantPartitionHelper out of delegates/gpu, and put into util.h as the logic remains same w/ other delegates that need to support FP16. — gardener / detail
82. Clarify CPU/GPU infeed error messages. — gardener / detail

#688 (May 18, 2020, 11:47:00 AM)

1. Add a compile flag (i.e. adding "--define tflite_with_xnnpack=true" when using bazel to build, which corresponds to defining macro TFLITE_BUILD_WITH_XNNPACK_DELEGATE) to apply XNNPACK delegate in TFLite runtime. — gardener / detail
2. Fix Core ML delegate framework's include — gardener / detail
3. Bumps llvm version — gardener / detail
4. Slightly optimize quantized add. — gardener / detail
5. Add optimized MatrixBatchVectorMultiplyAccumulate for asymmetric inputs for sse — gardener / detail
6. fix escape in Core ML header processing — gardener / detail
7. Update GraphDef version to 405. — gardener / detail
8. compat: Update forward compatibility horizon to 2020-05-18 — gardener / detail

#687 (May 17, 2020, 11:47:00 AM)

1. Simplify some tests — gardener / detail
2. Integrate LLVM at https://github.com/llvm/llvm-project/commit/7af0c8559b6d — gardener / detail
3. Add F64 Sqrt test. — gardener / detail
4. Update GraphDef version to 404. — gardener / detail
5. compat: Update forward compatibility horizon to 2020-05-17 — gardener / detail

#686 (May 16, 2020, 11:47:00 AM)

1. Fuse BN and Relu in mkl path — yang.sheng / detail
2. Copy exponential_avg_factor attr in grappler — yang.sheng / detail
3. enabling mkldnn threadpool build options — srinivasan.narayanamoorthy / detail
4. adding version number. — srinivasan.narayanamoorthy / detail
5. add bias to transpose_conv TESTs. — peng.sun / detail
6. review comments. — srinivasan.narayanamoorthy / detail
7. review changes. — srinivasan.narayanamoorthy / detail
8. update — yang.sheng / detail
9. Avoid overhead for creating executors if there is no change in execution mode. — gardener / detail
10. Rollback to investigate failure — gardener / detail
11. Update ops_history for bincount. — gardener / detail
12. Remove the unnecessary address-returning operator and lamda expression. — gardener / detail
13. Use fully-qualified std::string in TraceMe — gardener / detail
14. Avoid overhead for creating executors if there is no change in execution mode. — gardener / detail
15. Integrate LLVM at https://github.com/llvm/llvm-project/commit/9d4b4f344d8e — gardener / detail
16. Bump open source llvm revision to 9d4b4f344d8ea917e082cf58d66b71c0171e1650 — gardener / detail
17. [MLIR/XLA] Constant sinking to control flow regions. — gardener / detail
18. Fix TF2XLA's InitGraph for unused feeds. — gardener / detail
19. Create bot_config.yml — github / detail
20. Extend Keras Lambda layers to work with functions of any signature rather than only functions that take one argument. — gardener / detail
21. String formatting in assertAllEqual() fails in Python 3 because bytestring may — gardener / detail
22. [tfls.util] Remove tensorflow-lite-gpu from dependencies. — gardener / detail
23. [XLA:Python] Expose new use_spmd_partitioning compile option. — gardener / detail
24. Add dispatch support to more Python APIs. — gardener / detail
25. Update bot_config.yml — github / detail
26. Check for `_metrics` in case sublayer is resetting `_metrics` property. — gardener / detail
27. Update bot_config.yml — github / detail
28. Fix TFLite builds on Windows/MacOS — gardener / detail
29. Add NNAPI delegate support for Elu — gardener / detail
30. Attempt to build libtensorflow GPU with the manylinux2010 toolchain. — gardener / detail
31. Minor cleanup for strings in xplane_to_profile_response. — gardener / detail
32. Reduce Functional.__call__ Python overhead by ~5-10% — gardener / detail
33. Provide builtin_op_kernels target with Ruy and GEMV caching unconditionally enabled — gardener / detail
34. Skip TFE_ContextAsyncWait for tfrt. In current TF-TFRT integration, all ops are executed synchronously. We will revisit this later. — gardener / detail
35. When restoring a variable with an initializer, pass through restore metadata rather than forgetting it — gardener / detail
36. Delete Tensor constructor that takes a pointer. Otherwise, say, std::make_unique<Tensor>(GetTensorSomewhereThatActuallyReturnsAPointer()) would construct boolean tensor without a compile time error. — gardener / detail
37. Update Eigen to: https://gitlab.com/libeigen/eigen/-/commit/9b411757abd8458f9689b1384c6bf75da9b82357 — gardener / detail
38. Fix case where embedding column -> use_safe_embedding = false is used with variable partitioning. — gardener / detail
39. Clarify why we have *-gpu package. — gardener / detail
40. Change XLA's default to disable cpu_fast_math options with the exception of min_max behavior. — gardener / detail
41. Extend Keras Lambda layers to work with functions of any signature rather than only functions that take one argument. — gardener / detail
42. Add Unsupported dtype in tfrt for backward compatibility. We will use this dtype to support legacy types (e.g. DT_RESOURCE, DT_VARIANT) that are not natively implemented in TFRT. — gardener / detail
43. Create _HostComputeMlir for use in TF MLIR. — gardener / detail
44. Enable MLIR saved model import by default in TFLiteConverterV2's saved model API — gardener / detail
45. Move GetDeviceCoordinates() function and related constants in tpu_rewrite pass to common utility file. — gardener / detail
46. optimize for int8 add. — gardener / detail
47. Enable TextVectorization to be called on lists of strings (or lists of list-wrapped strings). Using NumPy arrays of characters is generally a bad practice because of their extreme memory usage. — gardener / detail
48. Implement NNAPI QoS APIs in NNAPI delegate. — gardener / detail
49. Add benchmarks for scalar conversions — gardener / detail
50. Support cancellation in multi-device and distributed function execution. — gardener / detail
51. Fixes linkage error on pre-18 Android where GLESv3 is not available — gardener / detail
52. Handle head extract outside compilation cluster device assignment. — gardener / detail
53. Enable SVD shape test in eager mode. — gardener / detail
54. Include what you use for the micro_framework bazel target. — gardener / detail
55. Remove debug message — gardener / detail
56. [tf.lite] Avoid designated initializers — gardener / detail
57. Print message about dumped MLIR modules — gardener / detail
58. Converting some of the dependencies in tensorflow/c/BUILD to use portable_tensorflow_lib_lite_no_runtime. — gardener / detail
59. added missing CHECK, whitespace cleanup — gardener / detail
60. Support int16 quantization type — gardener / detail
61. include-what-you-use for lite/core/api:api — gardener / detail
62. Add a TFE_Py_Execute traceme, which is the entrance to TF c++ — gardener / detail
63. Update (non-gradient) tests for tf.linalg.qr to also run in eager mode. — gardener / detail
64. Add TraceMeEncode helper — gardener / detail
65. Update TPU rewrite pass to populate replicated host devices on tf_device.replicate. — gardener / detail
66. Remove LazyLoader dependency from lite.interpreter — gardener / detail
67. Enable tf.linalg.matrix_solve tests in eager mode. — gardener / detail
68. Enable tf.linalg.matrix_solve_ls tests in eager mode. — gardener / detail
69. [XLA:GPU] [NFC] Add more logging output to explain fusion decisions — gardener / detail
70. Adding an option to tensor tracer to create a suffix folder based on the fingerprint of the tf.graph. — gardener / detail
71. [tf.data] Reduce verbosity of a warning as it is not actionable and appears for all programs that use tf.data with tf.distribute. — gardener / detail
72. [XLA] Calculate and print statistics about prefetches and evictions. — gardener / detail
73. Fix the quantile accumulator to always return at most the number of requested boundaries. — gardener / detail
74. Fix the quantile accumulator to always return at most the number of requested boundaries. — gardener / detail
75. [tfls.metadata] Use java7 version opts to build metadata lib. — gardener / detail
76. [tf.data] Update output time functions to solve the stack overflow problem. Also update some mathematics computation in the code. — gardener / detail
77. Update GraphDef version to 403. — gardener / detail
78. compat: Update forward compatibility horizon to 2020-05-16 — gardener / detail
79. Fix windows build. — gardener / detail
80. optimize for int8 add. — gardener / detail
81. Optimize trivial RealDiv ops — gardener / detail

#685 (May 15, 2020, 11:47:00 AM)

1. Removing unreachable return — anselm.hahn / detail
2. Rearange the return — anselm.hahn / detail
3. Prcandidate1 (#2) — github / detail
4. better error msg for incorrect WAV format chunk — joshua.richard.meyer / detail
5. Update download_and_extract.sh — github / detail
6. Add mlir_graph_optimization_pass.h header to pip wheel — yong.tang.github / detail
7. Remove Python2 badge in raspberry Pi into README — github / detail
8. Update protobuf-java to 3.9.2 — yong.tang.github / detail
9. Removes duplicate space from retracing warning message — github / detail
10. Generate cubin headers for bias_add and relu. — gardener / detail
11. Introduce a higher-level function handling in the tracing oriented unified API — gardener / detail
12. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
13. Clarify docstring: At EOF, GFile.readline() returns "". — gardener / detail
14. Automated g4 rollback of changelist 311477582. — gardener / detail
15. Comment typo fix. — gardener / detail
16. [TF MLIR SI] Don't constant fold, only consider result of folding — gardener / detail
17. Removing TensorHandleList Delete APIs, since TensorHandleList pointer is owned by ConcreteFunction. — gardener / detail
18. Resolve trivial aliases for portable TensorFlow targets. — gardener / detail
19. [XLA:Python] Cache the backend in xla_client_test. — gardener / detail
20. Do not silently ignore ptxas compilation failures. — gardener / detail
21. Internal change — gardener / detail
22. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
23. Fix style in `op_hint.py` to match formatting from Copybara. — gardener / detail
24. Add test for PromoteVarHandlesToArgs pass for testing users of tf.VarHandleOps (NFC). — gardener / detail
25. Generate MLIR ops for TPU Host/Device communication for outside compilation. — gardener / detail
26. Add support for setting up a TF_OutputList from the client and use it to build function with multiple results — gardener / detail
27. Set sparse FullyConnected op version properly. — gardener / detail
28. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
29. [XLA:CPU] Allow C64 and C128 types in Sort(). — gardener / detail
30. Update gather_op_test and unique_op_test to use subTest for easier debugging. — gardener / detail
31. [TF:TRT] Enable concatenation_test and biasadd_matmul_test for TAP. — gardener / detail
32. Only add sub_index to _Arg nodes. — gardener / detail
33. Use subTest on einsum_test to make errors easier to understand. — gardener / detail
34. Add Minimum and Maximum ops to Hexagon delegate for uint8/int8 — gardener / detail
35. Reinstall tf-estimator-nightly since regular installs might not work. — gardener / detail
36. Added BUILD rules for the micro-frontend TF op so that we can use it as a C++ library. — gardener / detail
37. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
38. Fix bazel TFLM compilation w/ TF_LITE_STATIC_MEMORY — gardener / detail
39. Change more libraries in third_party/tensorflow/c/BUILD to depend on portable_tensorflow_lib_lite_no_runtime to support effort to reduce dependencies on mobile. — gardener / detail
40. Instrument the number of times the MLIR-based TF Bridge is enabled. — gardener / detail
41. Report remote target in error messages for gRPC eager service requests. — gardener / detail
42. Update svd_op_test to run (non-gradient) tests in eager as well as graph mode. — gardener / detail
43. Build DynamicSlice and DynamicUpdateSlice ops with MlirHloBuilder — gardener / detail
44. Fix hardswish test for ubsan. — gardener / detail
45. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
46. Generate separate pod for Core ML delegate — gardener / detail
47. Split index_lookup into string_lookup and integer_lookup. — gardener / detail
48. Adds utility methods for storing SignatureDefs in the metadata table in the flatbuffer — gardener / detail
49. Rework kernel check for fully_connected. — gardener / detail
50. Use CompactTextString instead of String for generating ops. — gardener / detail
51. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
52. Check the minimum metadata parser version in the MetadataExtractor Java library. — gardener / detail
53. This is an internal change not visible to the public. — gardener / detail
54. Fix a memory leak. — gardener / detail
55. Optimize int8 broadcast min. — gardener / detail
56. Addresses https://github.com/tensorflow/tensorflow/issues/38694 by ensuring TensorShape v1/v2 backwards compatibility. — gardener / detail
57. Integrate LLVM at https://github.com/llvm/llvm-project/commit/dad2e92eaf53 — gardener / detail
58. Add a new DelegateUtils::InvokeWithCPUFallback API for handling delegate failures automatically in Invoke(). This is especially useful for NNAPI usage, as drivers may evict clients. — gardener / detail
59. Update GraphDef version to 402. — gardener / detail
60. compat: Update forward compatibility horizon to 2020-05-15 — gardener / detail
61. [XLA] Verify statically shaped result type in `xla_hlo.reshape` operation — gardener / detail
62. Remove default initialization in tflite::StatefulNnApiDelegate::Data struct. — gardener / detail
63. Allow index typed values in `hlo_scalars_to_dimension_tensor`. — gardener / detail
64. Integrate LLVM at https://github.com/llvm/llvm-project/commit/854f5f332af4 — gardener / detail

#684 (May 14, 2020, 11:47:00 AM)

1. Mark QuantizationSpec methods that don't change member data as constant — lukas.geiger94 / detail
2. Respect inference type in DefaultQuantParamsPass — lukas.geiger94 / detail
3. add __cudaInitModule to cudart_stub.cc — baarts / detail
4. address comments on commit fc58d58923534e461d735a9a8b460d2dc8691ae5 — baarts / detail
5. Fix the copy bug. — fbastien / detail
6. Add a test — fbastien / detail
7. NFC: rename a variable — fbastien / detail
8. [tflite] add fp16 support for evaluation tools — koansin.tan / detail
9. Update docstring of tf.split to clarify num_or_size_splits — yong.tang.github / detail
10. Combine into one block in example of docstring so that it could be rendered correctly — yong.tang.github / detail
11. Update docstring to remove confusing term num_split/size_split, based on review — yong.tang.github / detail
12. Fix the issue of tf.divide's return value is not a tensor — yong.tang.github / detail
13. add fp16 option to NNAPI delegate — koansin.tan / detail
14. add `nnapi_allow_fp16` option to benchmark_model — koansin.tan / detail
15. change and cleanup per review — koansin.tan / detail
16. fix bad indent and remove leftover — koansin.tan / detail
17. Change the default value of 'use_coreml' parameter to false — gardener / detail
18. Qualify uses of std::string — gardener / detail
19. Add SSE4 path for Tanh and Logistic. — gardener / detail
20. Integrate LLVM at https://github.com/llvm/llvm-project/commit/897d8ee5cd69 — gardener / detail
21. Bump open source llvm revision to 897d8ee5cd693e17f95a7e84194bca4c089a520b — gardener / detail
22. Separate out creating arguments from PromoteResourcesToArguments in PromoteResourcesToArgsPass (NFC). — gardener / detail
23. Fix batch_to_space's formatting. — gardener / detail
24. For python op generation: add dispatch to all generated ops (don't skip ops with VISIBILITY=HIDDEN) — gardener / detail
25. Add PromoteVarHandlesToArgsPass to lift tf.VarHandleOps from function bodies to function arguments. — gardener / detail
26. Add support for global operation dispatchers.  (This is intended for use by TF-internal classes only.) — gardener / detail
27. Disable failed test for now. — gardener / detail
28. Explain the reason to call _truediv_python3/_div_python2 explicitly (not through registered '/' — yong.tang.github / detail
29. Internal change — gardener / detail
30. Add support for global operation dispatchers.  (This is intended for use by TF-internal classes only.) — gardener / detail
31. Set up TFRT OSS dependency in Tensorflow. — gardener / detail
32. Adding TensorHandleList as the returned type for TF_ConcreteFunctionGetCaptures. — gardener / detail
33. Flush denormals to zero in eager mode. — gardener / detail
34. Use round-robin approach to reading from tf.data service workers. — gardener / detail
35. [XLA] Move SPMD partitioner to third_party — gardener / detail
36. Roll back change: "For python op generation: add dispatch to all generated ops (don't skip ops with VISIBILITY=HIDDEN)" — gardener / detail
37. Update tf.InplaceUpdate summary and description so both TensorFlow op registry and TensorFlow MLIR ODS match. — gardener / detail
38. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
39. Rolling change forward again: "Add support for global operation dispatchers.  (This is intended for use by TF-internal classes only.)" — gardener / detail
40. Simplify comment and use tensor_util.is_tensor and convert_to_tensor if both x and y are not tensor — yong.tang.github / detail
41. Remove the need to check if y is a tensor (always convert x if not) — yong.tang.github / detail
42. Fix Windows build after cl/311081931. — gardener / detail
43. Improve image testing for TFLite Java tests — gardener / detail
44. Move tf.keras.layers.featureDenseFeature back to Keras package. — gardener / detail
45. Export tf.keras.layers.experimental.preprocessing.CategoryCrossing layer. — gardener / detail
46. [XLA] Basic (R1) support for CPU bounded dynamic shapes. — gardener / detail
47. [tf.lite] Adds a setQuantizedModelsAllowed() Java API for running quant models with GPU delegate — gardener / detail
48. Make regularizers API more consistent. — gardener / detail
49. Support running a remote function with packed input handles. — gardener / detail
50. Fix issue where metric instances created in subclassed layer are not tracked on creation but only on call. — gardener / detail
51. Move TPUClusterResolver into tpu subdirectory. — gardener / detail
52. Correctly handle empty matrices in tf.linalg.svd. — gardener / detail
53. More robustly check for undefined symbols before attempting to use them. This check is required because undefined symbols are initialized with a special placeholder before entering control flow. This placeholder can lead to confusing error messages if left unchecked. The change introduces two more general  operators: "variable load" and "return". — gardener / detail
54. [Grappler] Do not add control edges from placeholder inputs in function inlining — gardener / detail
55. Export RandomZoom after its odd behavior was fixed. — gardener / detail
56. Make SerializeRemoteTensorHandle block only when the remote op is a function, in order to still benefit from async execution. — gardener / detail
57. Update attr name for Dense version Bincount. — gardener / detail
58. Register sparse FullyConnected kernel by default. — gardener / detail
59. [XLA:SPMD] Fix reshape with halo exchange — gardener / detail
60. Simplify `trainable` in batchnorm layers to just use the python variable rather than the private _trainable_var created by `backend.freezable_variable` — gardener / detail
61. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
62. Add simple canonicalizer for tf.fill — gardener / detail
63. Add a few more utility functions for TPUs — gardener / detail
64. Create per worker datasets in `strategy.experimental_distribute_datasets_from_function` instead of in `__iter__`. This will avoid tracing `dataset_fn` each time creating a new DistributedIterator. — gardener / detail
65. [tf.data] Update some maths formulas in the ComputeWaitTime function. — gardener / detail
66. Legalize TF broadcasting ops via dedicated xla_chlo.broadcast_* ops. — gardener / detail
67. Simplify error logging in XNNPACK delegate — gardener / detail
68. Internal change — gardener / detail
69. [XLA] Add use_names column to buffer info debug string to help with debugging. — gardener / detail
70. Reuse existing util functions in ReplicateToIslandPass and add back a test for remapping results (NFC). — gardener / detail
71. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
72. Remove nanopb dependency as it should not be needed anymore (grpc version on master no longer needs it) — gardener / detail
73. [XLA:SPMD] Avoid designated initializer. — gardener / detail
74. Fix the functional model loading with nested sequential model. — gardener / detail
75. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
76. Switch weights from per-value to per-input-item. — gardener / detail
77. support tpu strategy for crossing. — gardener / detail
78. Add the newly-added 'nnapi_allow_fp16' option to README. — gardener / detail
79. Optimize broadcast int8 max. — gardener / detail
80. Drop the dependency on ruy:detect_arm, which is becoming private. — gardener / detail
81. Legalize tflite CustomOp — gardener / detail
82. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
83. Update speech training notebook to use quantized inputs and outputs — gardener / detail
84. optimize for int8 add. — gardener / detail
85. Replace SameOperandsAndResultType by TFL_TCresVTEtIsSameAsOp to cover quantization types — gardener / detail
86. Internal change — gardener / detail
87. Pin keras_preprocessing to 1.1.0 — gardener / detail
88. Cleanup `setup.py` — gardener / detail
89. Change xtensa optimized softmax to use precomputed lookup table for quantized exponent calculation.  Use new memory API for softmax. — gardener / detail
90. Switch weights from per-value to per-input-item. — gardener / detail
91. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
92. Expose inference type in the mlir quantizer — gardener / detail
93. Add IOS_BENCHMARK enum value to BenchmarkType — gardener / detail
94. Fix comment to reflect actual logic. — gardener / detail
95. Return a meaningful error for dynamic shape inputs with outside compilation head extraction in TPUs. — gardener / detail
96. Integrate LLVM at https://github.com/llvm/llvm-project/commit/bfa200ebcf37 — gardener / detail
97. Update GraphDef version to 401. — gardener / detail
98. compat: Update forward compatibility horizon to 2020-05-14 — gardener / detail
99. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
100. Add Starlark rules to generate cubin headers. — gardener / detail
101. Bump open source llvm revision to bfa200ebcf3706fde0dde335a3c1fa3fe1b3ba3f — gardener / detail
102. Return a meaningful error for dynamic shape inputs with outside compilation head extraction in TPUs. — gardener / detail
103. Update XNNPACK dependency — gardener / detail

#683 (May 13, 2020, 11:47:00 AM)

1. [Lite] data: Fix memory leak — gaurav1086 / detail
2. Fix build errors — gaurav1086 / detail
3. [XLA] Adapting HLO-to-LHLO-legalization to use Buffer Assignment — ehsan.nadjaran_toosi / detail
4. fix dequantize op regression issue — leslie.fang / detail
5. Added POC for tf_program — cs17btech11050 / detail
6. Added no_oss_py2 tag and changed srcs_version for tests to PY3 — cs17btech11050 / detail
7. Refactoring: Format String -> Format Method — rladhkstn8 / detail
8. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
9. Add RaggedTensor dispatch for tf.nn.dropout. — gardener / detail
10. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
11. Bug fix for composite tensors in SavedModel — gardener / detail
12. Resolve trivial aliases for portable TensorFlow targets. — gardener / detail
13. Integrate LLVM at https://github.com/llvm/llvm-project/commit/123bee602a26 — gardener / detail
14. Minor fix an error in error message — gardener / detail
15. Bump open source llvm revision to 123bee602a260150ff55c74287f583a67ee78f36 — gardener / detail
16. Add support for a device ID op in parallel_device — gardener / detail
17. Resolve trivial aliases for portable TensorFlow targets. — gardener / detail
18. [XLA:Python] Make sure xla_client is always imported before TPU client extension. — gardener / detail
19. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
20. This breaks multi-python: — gardener / detail
21. IWYU in profiler/internal — gardener / detail
22. Add json translation for tfjs mlir converter. — gardener / detail
23. Install the `wrapt` pip package. — gardener / detail
24. Refactor MLIR TF shape inference to have a context — gardener / detail
25. Add bz2-devel so python will be compiled with bz2 support. — gardener / detail
26. Update manylinux docker image to latest hash. — gardener / detail
27. [tfdbg2] Ensure initialization on DebugEventsWriter.WriteGraphExecutionTrace() — gardener / detail
28. Implement outside compilation head extraction. — gardener / detail
29. [XLA/GPU] Make Thunk::Initialize() happen at compile-time, not run-time. This simplifies GpuExecutable for MLIR transition. — gardener / detail
30. Update hexagon_delegate guide to reference delegate readme — gardener / detail
31. Initial checkin of C++ header-only TensorHandle as part of RFC https://github.com/tensorflow/community/pull/207. — gardener / detail
32. Add Int8/BF16 1D un-tiled layout support for TPU. Host can transfer data — gardener / detail
33. Code cleanup: Use the combined left-or-right shift instruction. — gardener / detail
34. Add layout config to HloModuleConfig. — gardener / detail
35. [BUILD] Create a separate BUILD file for "tensorflow/core/protobuf"`. — gardener / detail
36. The callback slowness warning has started firing in many situations where only built-in callbacks are called (possibly only due to logging). For the time being, its threshold must be increased. — gardener / detail
37. Fix for adhering to latest clang style guide. — gardener / detail
38. Introduce persistent, read-only TFLite tensor type — gardener / detail
39. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
40. Update TensorFlow Lite external documentation — gardener / detail
41. Fix Resize Nearest Neighbor versioning. align_corners support was added in version 3. — gardener / detail
42. Make `core.Tensor` the base type for Tensor and replace the `register_dense_tensor_like` with direct subclassing. — gardener / detail
43. Disable collective ops xla test on gpu — gardener / detail
44. Make CHLO->HLO patterns extend OpRewritePattern vs OpConversionPattern. — gardener / detail
45. Add TF_AllTypesMatch trait in TensorFlow dialect. — gardener / detail
46. Modify op version in optimize only if convertor version < quantized version. — gardener / detail
47. [XLA/GPU] Make Thunk::Initialize() happen at compile-time, not run-time. This simplifies GpuExecutable for MLIR transition. — gardener / detail
48. Fix `alignment_hint < sizeof(void*)` in BuiltinDataAllocator::Allocate — gardener / detail
49. Update the OSS image to pickup latest tf estimator PIP package. — gardener / detail
50. [tf.data] Update the node destructor to solve the stack overflow problem. — gardener / detail
51. Enable Reduce Mean op in Hexagon delegate for uint8/int8. — gardener / detail
52. Refactor Softmax and use new memory API. — gardener / detail
53. Disable collective ops xla test on gpu — gardener / detail
54. Move tf.keras.layers.featureDenseFeature back to Keras package. — gardener / detail
55. Disable memory_optimizer_test on windows — gardener / detail
56. Restructure the Keras class hierarchy for Network, Model and Sequential. — gardener / detail
57. IWYU in profiler/lib — gardener / detail
58. Flush denormals to zero in eager mode. — gardener / detail
59. Add optional error reporting and status returns to MicroOpResolver — gardener / detail
60. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
61. Flush denormals to zero in eager mode. — gardener / detail
62. Validate remote resource devices before safe access of resources. — gardener / detail
63. Remove deprecated variants of DynamicSlice and DynamicUpdateSlice builders — gardener / detail
64. Support Core ML 3 for Core ML delegate when running on iOS 13 — gardener / detail
65. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
66. Updated image_test to work with newer versions of keras_preprocessing — gardener / detail
67. Support setting Core ML delegate's target Core ML version in benchmark — gardener / detail
68. Allow dynamically configuring device placement — gardener / detail
69. Disable flaky test — gardener / detail
70. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
71. Integrate LLVM at https://github.com/llvm/llvm-project/commit/58bc507b6fe6 — gardener / detail
72. Update GraphDef version to 400. — gardener / detail
73. compat: Update forward compatibility horizon to 2020-05-13 — gardener / detail
74. Bump open source llvm revision to 1c44430e738ba83eefe6d56a245ee30649d8988d — gardener / detail
75. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#682 (May 12, 2020, 11:47:00 AM)

1. Support Keras grouped convolutions — lukas.geiger94 / detail
2. Improve error messages — lukas.geiger94 / detail
3. Test group convolution error messages — lukas.geiger94 / detail
4. Test group convolution — lukas.geiger94 / detail
5. Allow groups=None — lukas.geiger94 / detail
6. Use test_util.use_gpu() — lukas.geiger94 / detail
7. Improve grouped convolution tests — lukas.geiger94 / detail
8. Update api/golden — lukas.geiger94 / detail
9. Fix group convolution docstring — lukas.geiger94 / detail
10. Improve group convolution docstring — lukas.geiger94 / detail
11. Provide NVIDIA CUDA build data in metadata and API — angerson / detail
12. Add NVIDIA CUDA and cuDNN info to tf.config — angerson / detail
13. Convert build_info to dict format and expose it. — angerson / detail
14. Fixed compilation errors with config=mkl — ag.ramesh / detail
15. Fix incorrect call to tf_export — angerson / detail
16. Added missing header file — ag.ramesh / detail
17. Remove unnecessary objc_library target — keithbsmiley / detail
18. Fix typos from the latest version — angerson / detail
19. Remove extra debug print statement — angerson / detail
20. Fixed circular dependency when building on Windows. — ag.ramesh / detail
21. Fix for UT failure in graph_runner_test — sharada.shiddibhavi / detail
22. Add weak frameworks back — keithbsmiley / detail
23. add -DDEBUG_BUILD to dbg profile — baarts / detail
24. Add to the cc_library too — keithbsmiley / detail
25. Fix bogomips extraction on s390x arch — rishi / detail
26. Increase the minimum support iOS version from 10 to 11. — gardener / detail
27. [XLA:CPU/GPU] Merge the elemental implementations of Reduce/ReduceWindow from CPU&GPU — gardener / detail
28. Shard a slow test — gardener / detail
29. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
30. [XLA:CPU/GPU] Merge the emission of elemental kMap — gardener / detail
31. Delete the useless print. — gardener / detail
32. Internal change only. — gardener / detail
33. Build clean up (OVIC). — gardener / detail
34. Support MirrorPad in hexagon — gardener / detail
35. Remove the special handling of TRTEngineOp in saved_model loader. — gardener / detail
36. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
37. Fix ANEURALNETWORKS_BAD_DATA execution failures with sum/max/min/reduce operations with scalar inputs. — gardener / detail
38. IWYU in profiler/utils — gardener / detail
39. Update tfrt hash in tf — gardener / detail
40. Fix typo in the test names: UnifedCAPI->UnifiedCAPI — gardener / detail
41. Export the `keras.preprocessing.image.smart_resize` utility. — gardener / detail
42. legalize operation 'tf.AllToAll'. Legalization target is xla_hlo.all_to_all. — gardener / detail
43. bump notebook versions — gardener / detail
44. [Profiler] Add optimization advices for tf-function. — gardener / detail
45. Refactor ConvertToTensorProto to avoid some duplication — gardener / detail
46. Added Pad3DAttributes attributes and function for shape calculation with Pad3DAttributes. — gardener / detail
47. Split tensorflow.python.tpu module doctests into different target. — gardener / detail
48. Build _TPUCompileMlir and TPUCompileSucceededAssert ops outside of parallel_execute region. — gardener / detail
49. Simplify tf::ConstOp's inferReturnTypes — gardener / detail
50. Merge TPU execution device with host device (NFC). — gardener / detail
51. Reduce space required for TFLM by dropping two non-needed pointers from MicroAllocator. — gardener / detail
52. Re-update the TFRT hash. — gardener / detail
53. Reduce inference functions exposed — gardener / detail
54. Remove disable_tfrt for scalars are not handled reason, as scalars are handled properly with TFRT. — gardener / detail
55. Integrate LLVM at https://github.com/llvm/llvm-project/commit/728cf6d86b4f — gardener / detail
56. Add tpu_strategy into TPU doctests. — gardener / detail
57. Lowering InplaceUpdate to DynamicUpdateSlice — gardener / detail
58. Bump open source llvm revision to 728cf6d86b4f20144ac10517afb0cb978beac124 — gardener / detail
59. Fix build failure in windows — gardener / detail
60. ReverseV2 lowering to xla_hlo.reverse — gardener / detail
61. Adjust the tool data for OSS TF memory profiler. — gardener / detail
62. [tf.data service] Fix max_outstanding_requests. — gardener / detail
63. IWYU in profiler/convert — gardener / detail
64. Internal micro demo. — gardener / detail
65. Mark multi-process utilities with subprocess module as deprecated in favor of using MultiProcessRunner. — gardener / detail
66. Update Materialize Broadcasts for same-rank broadcasts — gardener / detail
67. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
68. allow early termination of Profile Rpc for programmatic mode. — gardener / detail
69. Fix an ASAN failure due to uninitialized RemoteMgr. — gardener / detail
70. Internal change — gardener / detail
71. Fix benchmark naming in reporting — gardener / detail
72. Change default Grappler timeout from 5 to 20 minutes. Despite Grappler having become faster, graphs sizes have grown faster still, and the original default timeout is exceeded too frequently in some applications. — gardener / detail
73. Add support for lowering convert op to Linalg from floating-point to integer. — gardener / detail
74. Internal change — gardener / detail
75. [XLA] Expose SPMD/MPMD as an option — gardener / detail
76. Update the TensorFlow version properly. — gardener / detail
77. Change a couple bad variable types. — gardener / detail
78. Convert complex and unsigned integer tensors to and from dense elements attr — gardener / detail
79. Auto-generate ConjugateTranpose TensorFlow op — gardener / detail
80. Enable compiler tests disabled because of incomplete complex types support — gardener / detail
81. Fix Data type class/method naming collision. — gardener / detail
82. Release the GIL in file_io operations and reacquire it for error handling or Python specific operations. — gardener / detail
83. Add a TF-internal visibility declaration for the parallel device — gardener / detail
84. Fix the ImageDataGenerator methods to return Keras sequences instead of just generators. This makes it so that Keras `fit` avoids an infinite loop when users pass the results of ImageDataGenerator.flow* directly to fit/evaluate/predict. — gardener / detail
85. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
86. [tfls.release] Enable jar merging for bazel build of metadata lib. — gardener / detail
87. IWYU in profiler/rpc — gardener / detail
88. Deprecate `set_learning_phase()` backend utility. — gardener / detail
89. [tf.data] Add comment to ShouldWait for the nondeterministic case to explain the condition it's checking for — gardener / detail
90. Enable tests blocked by b/117943489 — gardener / detail
91. Add utilities for creating delegates in TF Lite — gardener / detail
92. tf.signal: Remove set_shape from documentation for linear_to_mel_weight_matrix. — gardener / detail
93. fix a typo — gardener / detail
94. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
95. Handle compilation of ops with dynamic shapes in fallback path — gardener / detail
96. Lower TF XlaConv op using the fallback lowering pass — gardener / detail
97. Whitelist XlaBroadcastHelperOp and enable tests XlaDynamicUpdateSliceOp — gardener / detail
98. Fix a comment about device/op_device. — gardener / detail
99. [tfls.image] Provide TensorImage#getWidth and TensorImage#getHeight as they are not trivial. — gardener / detail
100. Retrieve CUDA targets to build in nvcc wrapper from clang command line option. — gardener / detail
101. Add TPU Configuration C API — gardener / detail
102. Update GraphDef version to 399. — gardener / detail
103. compat: Update forward compatibility horizon to 2020-05-12 — gardener / detail
104. Use uint8_t, uint32_t and uint32_t. — gardener / detail
105. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
106. Factor out a virtual function for creating the TFLite interpreter so that subclass could create a customized one. — gardener / detail

#681 (May 11, 2020, 11:47:00 AM)

1. Integrate LLVM at https://github.com/llvm/llvm-project/commit/c5e0967e4cf0 — gardener / detail
2. Bump open source llvm revision to c5e0967e4cf0f1337bec772949e6cede4c01354b — gardener / detail
3. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
4. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
5. Fix for Micro make patching error on MacOS — gardener / detail
6. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
7. Refactor ConvertToTensorProto to avoid some duplication — gardener / detail
8. Refactor ConvertToTensorProto to avoid some duplication — gardener / detail
9. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
10. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
11. Support align_corners and half_pixel_centers for resize ops in NNAPI delegate. — gardener / detail
12. Log the rendezvous address for debugging purpose. This is very helpful for — gardener / detail
13. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
14. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
15. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
16. Update GraphDef version to 398. — gardener / detail
17. compat: Update forward compatibility horizon to 2020-05-11 — gardener / detail
18. Use int32_t and uint32_t instead of int32 and uint32. — gardener / detail
19. Tolerate differences equal to `tolerated` threshold in MinMaxApproximatelyEqual. — gardener / detail
20. Install auditwheel in the multi-python container and move patchelf installation — gardener / detail
21. Bump open source LLVM revision to f40fc7b8d60e6897fd9514a46a26f4b84f57577a — gardener / detail

#680 (May 10, 2020, 11:47:00 AM)

1. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
2. Enable MLIR saved model import by default in TFLiteConverterV2's saved model API — gardener / detail
3. [XLA] Introducing AllGather HLO and a decomposition pass — gardener / detail
4. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
5. Fix the myth from height vs width. The cause is due to — gardener / detail
6. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
7. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
8. Update GraphDef version to 397. — gardener / detail
9. compat: Update forward compatibility horizon to 2020-05-10 — gardener / detail
10. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#679 (May 9, 2020, 11:47:00 AM)

1. General case of logistic/tanh, 16-bit -> 16-bit. — elena.zhelezina / detail
2. Versioning for activation functions Tanh/Sigmoid. — elena.zhelezina / detail
3. Addressed review comments. — elena.zhelezina / detail
4. Pydoc fixes for StructuredTensor — cutlerb / detail
5. Added field values to __repr__, fixed pydoc outputs — cutlerb / detail
6. Added 16x8 LeakyRelu kernel. — elena.zhelezina / detail
7. Updated versioning for LeakyRelu. — elena.zhelezina / detail
8. Fix tests that use __repr__ — cutlerb / detail
9. Fix lint error — cutlerb / detail
10. Corrected test for LeakyRelu. — elena.zhelezina / detail
11. Minor docstring change: delete "normal". — gardener / detail
12. [XLA:CPU] Fuse reduce-window — gardener / detail
13. [XLA:CPU] Fuse reduce-window — gardener / detail
14. Reuse the rendezvous provided by the OpKernelContext for PartitionedCallOp. — gardener / detail
15. Fold iOS-specific targets into the portable targets. — gardener / detail
16. configure.py: Simplify bazel version check — gardener / detail
17. [Executor] Avoid unnecessary `NodeItem::input_type()` calls in `PrepareInputs()`. — gardener / detail
18. Replace ptr_util.h with absl/memory/memory.h in the profiler — gardener / detail
19. Adding a FromBuffer method to construct tensorflow::cc::Tensors from user provided buffers. This also unblocks adding unit tests to tensorflow::cc::Tensor. — gardener / detail
20. Replace str_util.h with absl/strings in profiler — gardener / detail
21. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
22. Set the AssignedDevice of the lowered control flow nodes (Enter, Switch, Merge, NextIteration, Exit) to be the same as the assigned device of the input loop variable if that is available during lowering. — gardener / detail
23. Internal visibility addition. — gardener / detail
24. Remove use of gtl/cleanup.h and hash.h in cupti_tracer — gardener / detail
25. Add skip_if_error context manager in test_util.py to conveniently skip errors that are not related to what is being tested. — gardener / detail
26. Add a nested tf.function with control flow test. — gardener / detail
27. Resubmitting the change to make iterator deleter not on host memory — gardener / detail
28. Move c_api_tfrt to core/tfrt/eager (NFC) — gardener / detail
29. Reduce space required for TFLM by dropping two non-needed pointers from MicroInterpreter. — gardener / detail
30. Merge release notes and version updates to master — gardener / detail
31. Create Ubuntu 16 scripts and configs to resurrect libtensorflow releases. — gardener / detail
32. Change the logging from warning to error when there is an exception at session close. — gardener / detail
33. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
34. Add a comment to note that the structs are "semi-ABI stable" in TFL common.h. — gardener / detail
35. Branch LandmarksToTransformMatrix operation to LandmarksToTransformMatrixV2. — gardener / detail
36. Support running a function with packed input handles through C APIs. — gardener / detail
37. Use protobuf.h instead of proto_serialization.h in profiler — gardener / detail
38. Replace lib/core/{errors,status}.h with platform/ headers in profiler — gardener / detail
39. Fix requested_device of NextIteration node. — gardener / detail
40. Add support for quantizing prelu, now that its a builtin operator. It has two inputs. First input must match output scale. — gardener / detail
41. Cleanup headers in capture_profile.cc — gardener / detail
42. pfor: change converters for Equal and NotEqual to handle — gardener / detail
43. Replace processing of `tf_device.launch` with `tf_device.cluster` in TPUExtractOutsideCompilation pass. — gardener / detail
44. Cleanup headers in save_profile.cc — gardener / detail
45. Convert speech training to produce an int8-quantized model — gardener / detail
46. [tf.lite] Fix issue with direct ByteBuffer inputs and dynamic graphs — gardener / detail
47. Merge portable full/lite proto libraries. — gardener / detail
48. Fix RandomZoom layer for outputing 0 values. — gardener / detail
49. [Executor] Fix segfault when using verbose logging in the executor module. — gardener / detail
50. Remove the special handling of TRTEngineOp in saved_model loader. — gardener / detail
51. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
52. Use subTest to improve error reporting on linear algebra ops. — gardener / detail
53. Remove `_enable_legacy_iterators` flag in TPUStrategy. — gardener / detail
54. Implement Numpy to tensor conversion for TFRT. — gardener / detail
55. Remove timing code from profiler — gardener / detail
56. Remove unused header — gardener / detail
57. Import and export constants of complex types for HLO — gardener / detail
58. Enable memory_space verification for fusion input parameters. — gardener / detail
59. Add missing ASSERT_EQ on status after API call in c_api_unified_experimental_test — gardener / detail
60. Split table management off into a table_utils file. — gardener / detail
61. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
62. Introduce some common constants for TPU. — gardener / detail
63. Allow kernels to take different scales for prelu — gardener / detail
64. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
65. Remove redundant negations in HiFi mini's fully_connected implementation. — gardener / detail
66. Code cleanup: Use the combined left-or-right shift instruction. — gardener / detail
67. Make tf.While work with ConcreteFunction. — gardener / detail
68. Add NDEBUG to default make build. — gardener / detail
69. Allow Op version of 0 to match the first registered Op of that type. — gardener / detail
70. Internal change — gardener / detail
71. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
72. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
73. compat: Update forward compatibility horizon to 2020-05-09 — gardener / detail
74. Update GraphDef version to 396. — gardener / detail
75. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#678 (May 8, 2020, 11:47:00 AM)

1. Fix document error for ApplyFtrl, ApplyFtrlV2 and SparseApplyFtrlV2 — blueyiniu / detail
2. Support options(environment variable) to enable grpc reuse port. — tongxuan.ltx / detail
3. fix typo — tongxuan.ltx / detail
4. fix build break — tongxuan.ltx / detail
5. Fix misspelling — mshr-h / detail
6. Add cardinality calculation for Dataset.unbatch() when possible — yong.tang.github / detail
7. Add test case for cardinality with Dataset.unbatch() — yong.tang.github / detail
8. Update based on review feedback. — yong.tang.github / detail
9. Add additional test case where only the second batch size is known (from the review comment) — yong.tang.github / detail
10. Check return status of reading environment variable — root / detail
11. Enable Squeeze Op conversion without squeeze_dim attribute in explicit batch mode. — tfeher / detail
12. Let the test use the new ruy::ReferenceMul. — gardener / detail
13. Add  "//tensorflow/compiler/mlir/tensorflow:graph_optimization_pass_registration" to tensorflow/python/_pywrap_mlir.so — yong.tang.github / detail
14. [XLA] Add a memory space propagation pass. — gardener / detail
15. Add named size and count methods for arg, result and var methods to AOT models. — gardener / detail
16. Export the only one function of a saved model only when it matches with exported_names argument. — gardener / detail
17. Enable XRT cache to be shared among multiple GPU devices. — gardener / detail
18. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
19. Fix control dependency issue causing shape_assert test to fail. — gardener / detail
20. Emit error messages for all missing legalizations in TF to XLA full legalization pass. — gardener / detail
21. Delete duplicate installation of patchelf. — gardener / detail
22. Test ConvertUnary in dynamic shape mode — tfeher / detail
23. Use persistent buffer in quantize, dequantize and xtensa_hifimini/quantize. — gardener / detail
24. Fix output shape of tf.GatherV2 in legalize-tf.mlir — gardener / detail
25. Use subTest to improve error reporting on test failures in cwise_ops_test.py. — gardener / detail
26. Branch TransformLandmarks operation to TransformLandmarksV2. — gardener / detail
27. [XLA] Correct WhileLoopInvariantCodeMotion log messages. — gardener / detail
28. Internal changes only. — gardener / detail
29. Use multiple async executor to launch collectives in eager instead of tf.function — gardener / detail
30. Update Eigen to: https://gitlab.com/libeigen/eigen/-/commit/49f1aeb60d9f759859fce0d16aa5d1ecc7168d51 — gardener / detail
31. [ROCm] Fix for ROCm CSB breakage on 200507 — deven.desai.amd / detail
32. Use strategy default device (if not None) as the destination for strategy.reduce. — gardener / detail
33. Remove multi_worker_callback_tf1_test for lack of resource to continue maintenance with the focus on tf2/eager. — gardener / detail
34. Whitelist a few more unary and binary TensorFlow ops for the fallback path — gardener / detail
35. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
36. Implement fast deferred-decoding Python stack trace class. — gardener / detail
37. Add util function to determine CPU host device from TPU device. — gardener / detail
38. Static link GPU OpenGL tests — terryheo / detail
39. Implement fast deferred-decoding Python stack trace class. — gardener / detail
40. Fix documentation compatibility tag formatting. — gardener / detail
41. Disable a test from sanitizers. — gardener / detail
42. [tf.data] Bug fix: make_csv_dataset should not modify mutable parameters passed into it — gardener / detail
43. This will break once kwargs has no key of 'partitioner'. — gardener / detail
44. Enable MLIR bridge for ops that are already supported — gardener / detail
45. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
46. [Executor] Optimize `PropagatorState::FindOrCreateChildFrame()`. — gardener / detail
47. Don't concatenate empty tensors — gardener / detail
48. Integrate LLVM at https://github.com/llvm/llvm-project/commit/910871532101 — gardener / detail
49. Make regularizers API more consistent. — gardener / detail
50. Try https and see if that works. — gardener / detail
51. CalculateOutputShape for concatenation of BHWDC tensors. — gardener / detail
52. Fix enable_v2_dtype_behavior() doctest failure. — gardener / detail
53. Internal change — gardener / detail
54. Add op sanity checks to the following TFLite ops: — gardener / detail
55. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
56. Enable Keras/RNN model via conversion routes that accept a keras model. — gardener / detail
57. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
58. Add i386 to ObjC podspec — gardener / detail
59. Add a kernel generator tool. — gardener / detail
60. Update GraphDef version to 395. — gardener / detail
61. compat: Update forward compatibility horizon to 2020-05-08 — gardener / detail
62. Internal BUILD file change. — gardener / detail
63. [XLA][MLIR] Lower complex operations to std dialect. — gardener / detail

#677 (May 7, 2020, 11:47:00 AM)

1. update dataset_ops.py — github / detail
2. Shorten/wrapped the line 1659 — github / detail
3. added parentheses on line no 1659 to avoid syntax error. — github / detail
4. Update cusparse_10_1.inc — nluehr / detail
5. Extend sparse kernels.h API with workspace size — nluehr / detail
6. Use csrgeam2 for CUDA 10.0+ — nluehr / detail
7. Replace csrgemm with csrgemm2 for CUDA 10.0+ — nluehr / detail
8. Replace cusparseCsr2csc with cusparseCsr2cscEx2 for CUDA 10.1+ — nluehr / detail
9. Replace Csrmm with SpMM for CUDA 10.2+ — nluehr / detail
10. Replace csrmv_mp and csrmv with csrmvEx and SpMV for CUDA 10.2+ — nluehr / detail
11. Bumped up versions for 16x8 reference kernels. — elena.zhelezina / detail
12. Tidy up. — elena.zhelezina / detail
13. Fix exception handling of tf.histogram_fixed_width_bins — yong.tang.github / detail
14. Add test case for exception handling in `tf.histogram_fixed_width_bins` — yong.tang.github / detail
15. Pylint fix — yong.tang.github / detail
16. Added ... to add multiline python statement. — github / detail
17. Revert "Update cusparse_10_1.inc" — nluehr / detail
18. Add windows support for s3 — vovannghia2409 / detail
19. Fix s3 test on Windows — vovannghia2409 / detail
20. Delete args, fix HasAtomicMove — vovannghia2409 / detail
21. Change to use tensor_util.constant_value if possible and raise ValueError — yong.tang.github / detail
22. Explicit TensorShape in allocate_temp — nluehr / detail
23. Cleanup formatting with clang-format — nluehr / detail
24. Fix Ubuntu Sanity — vovannghia2409 / detail
25. Added TANH op and test to micro — yair.ehrenwald / detail
26. Pylint fix to remove unused import — yong.tang.github / detail
27. Bump open source llvm revision to 307cfdf5338641e3a895857ef02dc9da35cd0eb6 — gardener / detail
28. make `bazel build` work on aarch64/arm linux — koansin.tan / detail
29. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
30. Move "-c opt" before "--config" just for being consistent. — gardener / detail
31. Disable flaky XLA test on GPU — gardener / detail
32. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
33. [tfdbg2] Implement the to_json() methods of two data classes — gardener / detail
34. Improve TensorFlow arithmatic op folders — gardener / detail
35. [ArgOp] Use `CanConsumeArg()` and `ConsumeArg()` when possible. — gardener / detail
36. [WhileOp] Support forwarded inputs in the `ComputeAsync()` path. — gardener / detail
37. Disable flaky test — gardener / detail
38. Include new op EnqueueTPUEmbeddingRaggedTensorBatch in the list of TPUEmbedding enqueue ops don't get auto control edges created so that they can be used in TF2/eager mode in tf.functions. — gardener / detail
39. Internal change — gardener / detail
40. Replace safeguard against aliasing cond variables with a dynamic check. It is more verbose but more robust in the presence of exception-driven control flow. — gardener / detail
41. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
42. Some tests fixes. — gardener / detail
43. Disable only the TPU version of distribute_strategy_test — gardener / detail
44. [PropagatorState] Use `IterationState*` instead of `int64` iteration ID in `TaggedNode`. — gardener / detail
45. Fix layout_optimizer_test on Windows. — gardener / detail
46. Raises a more meaningful error when fdef is missing — gardener / detail
47. Improve TensorFlow arithmatic op folders — gardener / detail
48. Expose disable_per_channel in MLIR to be used experimentally by tflite tooling — gardener / detail
49. Add a script to install patchelf on docker containers and add it to multipython. Install auditwheel as well. — gardener / detail
50. Update make_tensor_proto to support nested nparray values when dtype is specified. — gardener / detail
51. Remove op_gen_lib and distributed runtime dependency for Cast. — gardener / detail
52. don't use StrValue() because in future, it can be saved in references. (string tables). — gardener / detail
53. Fix ROCm build errors caused by https://github.com/tensorflow/tensorflow/pull/38802. — gardener / detail
54. Update MLIR bridge passes to use `tf_device.cluster`/`tf_device.cluster_func` instead of `tf_device.launch`/`tf_device.launch_func` for defining cluster/computation. — gardener / detail
55. Update ruy to get ReferenceMul. — gardener / detail
56. Add ExpandDims and ConcatV2 ops to tf_to_corert lowering. — gardener / detail
57. Remove unused variable, and replace reinterpret_casts with static_casts. — gardener / detail
58. Add bincount to support dense/sparse/ragged inputs. — gardener / detail
59. string stats use reference value in tfstreamz plane. — gardener / detail
60. Sparse fully connected kernel with 1x4 block config. — gardener / detail
61. add a xplane builder test to test reference value etc. — gardener / detail
62. Add shell function to create a Bazel test summary XML file. — gardener / detail
63. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
64. Roll forward after sending out a fix for the breaking test — gardener / detail
65. Enable memory profile tracing by default in BFCAllocator. — gardener / detail
66. [XLA:Python] Remove xla_client.transfer_to_infeed and transfer_from_outfeed. — gardener / detail
67. use reference value for annotation/kernel details xstat. — gardener / detail
68. Re-arrange TFRT SavedModel directory. — gardener / detail
69. Add test to cover the code path run by the public multi-worker with Keras — gardener / detail
70. Don't put iterator deleters as host memory. — gardener / detail
71. Update index_lookup to return a UTF-8 string in vocabulary lookup instead of a bytestring. — gardener / detail
72. Don't put iterator deleters as host memory. — gardener / detail
73. In AggregatingVariable.assign, when aggregating values across replicas in a merge_call, also check if the "name" kwarg is a PerReplica object and if so, change to a single value. — gardener / detail
74. Fixes: GitHub Pull Request #37017 — gardener / detail
75. [XLA:Python] Remove xla_client.Buffer class. — gardener / detail
76. Disable broken Windows test — gardener / detail
77. Fix ConvertHalfTensor bug with elements attribute of size one — gardener / detail
78. Auto-generate some of the TensorFlow ops — gardener / detail
79. Update ops-related pbtxt files. — gardener / detail
80. Fixed int64 vs int32 issue in empty op folder. — gardener / detail
81. Fix an issue of out of order execution. Don't serialize a remote input handle for function execution until it's ready on a remote device. Otherwise, on a remote worker, a remote function execution request could be enqueued before a request for producing a function input. — gardener / detail
82. Fixes: GitHub Issue #39222 — gardener / detail
83. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
84. Emit error messages for all missing legalizations in TF to XLA full legalization pass. — gardener / detail
85. [tf.data] Add map and batch fusion rewrite in MLIR — gardener / detail
86. Internal test infra change — gardener / detail
87. Run shape inference before any new non-tf ops are introduced — gardener / detail
88. Update Hello World example. — gardener / detail
89. Increase test coverage for CentralStorageStrategy with 1 cpu and 1 gpu case — gardener / detail
90. Support unknown dimensions in quantized models. — gardener / detail
91. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
92. TFLM: Print out the buffer usage. — gardener / detail
93. Internal change — gardener / detail
94. Add more dtypes to scatter nd ops — gardener / detail
95. Fix ObjC podspec template — gardener / detail
96. Exclude MemoryAllocation and MemoryDeallocation TraceMes from the TF profiler Trace Viewer. — gardener / detail
97. Correct the C API path for the TFLite ObjC podspecs — gardener / detail
98. Podspec updates for TFLite v2.2.0 — gardener / detail
99. Automated refactoring to make code Python 3 compatible. — gardener / detail
100. Fix the real multiplier calculation — gardener / detail
101. Split merged TFLiteConverter implementations into frozen graph converter and saved model converter — gardener / detail
102. Migrate Model Maker tutorials to tf.org/lite — gardener / detail
103. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
104. Update GraphDef version to 394. — gardener / detail
105. compat: Update forward compatibility horizon to 2020-05-07 — gardener / detail
106. Fix liveness analysis for variables closed over by functions. Previously, these variables were only live at the closing function's point of definition. After this change, these variables are live at every statement that may be reached by this function definition. This is more consistent with Python's late binding mechanism. — gardener / detail

#676 (May 6, 2020, 11:47:00 AM)

1. [ROCm] Adding no_rocm tag to tests currently failing on ROCm platform — deven.desai.amd / detail
2. Spellcheck @keras.layers.convolutional.py — github / detail
3. Spellcheck @keras.layers.core.py — github / detail
4. Fix Huber Loss crashes due to data type mismatch — yong.tang.github / detail
5. Add test case for Huber Loss crashes when data type mismatch — yong.tang.github / detail
6. Fix BatchNormalization issue with virtual_batch_size when shape has None — yong.tang.github / detail
7. Add test case for BatchNormalization with virtual_batch_size and shape has None. — yong.tang.github / detail
8. Enable explicit batch mode transpose — tfeher / detail
9. Test ConvertTranspose in dynamic shape mode — tfeher / detail
10. Improve parameterization of tests — tfeher / detail
11. Use testing::Matcher in BuildAndRun, correcty style — tfeher / detail
12. Edit docstring of AddTestTensor — tfeher / detail
13. Change matcher type — tfeher / detail
14. Edit docstring — tfeher / detail
15. Fix issue in boolean_mask when axis is passed as a tensor — yong.tang.github / detail
16. Add test case for boolean_mask with axis passed as a tensor — yong.tang.github / detail
17. Extend support of non-const axis tensor in boolean_mask — yong.tang.github / detail
18. Add test case for non-const axis tensor in boolean_mask — yong.tang.github / detail
19. Address review comment (change test name and add URL) — yong.tang.github / detail
20. Update to use tf.shape to get the shape of the tensor, from review comment — yong.tang.github / detail
21. Improve comments and add a static cast — tfeher / detail
22. Resubmit of PR 38848 but only support complex64 and complex128 — yong.tang.github / detail
23. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
24. [MLIR] Use dim expr for operand dim == 1 && result dim == 1 for BroadcastInDim. — gardener / detail
25. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
26. [MLIR] Reformat lhlo-legalize-to-linalg.mlir test. — gardener / detail
27. Update criteria for TPU job/worker experiments — gardener / detail
28. Increase Apache Spark version to 2.4.5 to handle GitHub Security Alert — gardener / detail
29. Qualify uses of std::string — gardener / detail
30. Replacing ::Cord with absl::Cord. — gardener / detail
31. Disable flaky test — gardener / detail
32. Make tranpose_op_test.py use subTest for clearer test debugging. — gardener / detail
33. Re-enabling optimizer_v2 test on OSS as the dependent bug has been closed. — gardener / detail
34. Edit docstring of ParameterizedOpconverterTest — tfeher / detail
35. Check that "VariableV2Op" doesn't exist in a TFLite-specific pipeline. — gardener / detail
36. Disable two TPU OSS tests. — gardener / detail
37. [CallFrameInterface] Add methods `ConsumeArg()` and `CanConsumeArg()`. — gardener / detail
38. Internal visibility change — gardener / detail
39. Enable tool data saving for OSS TF memory profiler. — gardener / detail
40. [XLA:CPU] Add our own __powisf2 and __powidf2 which can be generated by LLVM — gardener / detail
41. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
42. Forwardprop: opt the forwardprop utility function out of run_functions_eagerly — gardener / detail
43. Add lowering from xla_hlo.sine to linalg. — gardener / detail
44. Add macros to avoid frequent ROCm ifdefs. Clean up the ifdefs in core/kernels/sparse. — gardener / detail
45. Re-enabling keras_rnn_correctness_test for non tpu platforms as only the tpu version fails. — gardener / detail
46. There is a bug in the outer partition code in StructuredTensor for a dense field. — gardener / detail
47. Reorganizing kernel fallback into its own (kernel/) directory. Right now this looks a bit awkward, but long term organization will look as follows: — gardener / detail
48. Perform uint8<->int8 casting on DSP — gardener / detail
49. Fix typo in metadata_schema.fbs — gardener / detail
50. Deprecate Network `state_updates` property. — gardener / detail
51. Sort control edges on input considering src node — gardener / detail
52. filter out tfstreamz plane when generate trace events. — gardener / detail
53. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
54. Adds support for half_pixel_centers in TFLite's resize nearest neighbor op. — gardener / detail
55. [tf.data service] Add support for shared job names. — gardener / detail
56. Deflake cross_device_ops test. — gardener / detail
57. Support XlaCompiler arguments of type kConstant in XlaCompilationCache — gardener / detail
58. Fix legalizeTF(). Currently definition of this function is inside an anonymous namespace. — gardener / detail
59. Clean up code in the TF FixedPoint extension to Eigen to make the AVX2/AVX512 compile with recent versions of clang. — gardener / detail
60. Move AllPassesAndDialects registrations in flatbuffer converters to binary and test rules. — gardener / detail
61. Update array_ops_test to use subTest for easier debugging. — gardener / detail
62. Rollback: Not able to reproduce the failure. — gardener / detail
63. Update ops-related pbtxt files. — gardener / detail
64. Add auto generated TensorFlow op BroadcastArgs — gardener / detail
65. Enable MLIR bridge for compiler tests requiring unsigned integers — gardener / detail
66. Enable MLIR bridge for compiler tests that are already passing or requires whitelisting — gardener / detail
67. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
68. Add doc on building TFLite C++ libraries — gardener / detail
69. Fix bug in complex einsum grad. — gardener / detail
70. TFLite binary size optimization by removing templates — gardener / detail
71. [MLIR] Expose HLO -> LHLO transformation for XLA/GPU later use — gardener / detail
72. Fix row sums size calculation in lstm eval — gardener / detail
73. [XLA:Python] Update Device.transfer_from_outfeed to set outfeed layout if missing. — gardener / detail
74. Enable index_lookup to work with non-batched byte strings. — gardener / detail
75. When checking whether ExpandDims op has a constant dim value, also use m_Constant to match in case it's a mlir constant. — gardener / detail
76. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
77. When checking whether ExpandDims op has a constant dim value, also use m_Constant to match in case it's a mlir constant. — gardener / detail
78. Support MediaPipe pooling operators in XNNPACK delegate — gardener / detail
79. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
80. compat: Update forward compatibility horizon to 2020-05-06 — gardener / detail
81. Update GraphDef version to 393. — gardener / detail
82. Remove all uses of `TENSORFLOW_LITE_PROTOS`. — gardener / detail
83. Integrate LLVM at https://github.com/llvm/llvm-project/commit/307cfdf53386 — gardener / detail
84. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#675 (May 5, 2020, 11:47:00 AM)

1. Fixed int vs. int32_t mismatch for portability. — andrewc / detail
2. Fix test to run on all filesystem — vovannghia2409 / detail
3. Fix `fname` argument for `GetFileSystemForFile()` — vovannghia2409 / detail
4. Use TF_CALL_GPU_ALL_TYPES for variable ops — yong.tang.github / detail
5. Expand additional test to variable types for complex64 and complex128. — yong.tang.github / detail
6. Add needed dtype conversion, and pylint fix — yong.tang.github / detail
7. Change temp directory to work with Windows — vovannghia2409 / detail
8. Fix typo and run clang-tidy — vovannghia2409 / detail
9. [ROCm] Disabling a subtest within //tensorflow/python/eager:function_test_gpu because it checks output which is non-determnisitic in nature and therefore sporadically fails — deven.desai.amd / detail
10. Add tmp_dir — vovannghia2409 / detail
11. Get MACHTYPE via bash call instead of relying on the environment. — gardener / detail
12. Add flags to configure whether to fallback to the driver for ptx -> SASS. — gardener / detail
13. Add support for more LHLO ops in the MLIR emitter — gardener / detail
14. Added accessor for mutating instruction dimensions. — gardener / detail
15. Do not fail if local mirror already exists — gardener / detail
16. [tf.data] Cleanup after `cardinality` rollback. — gardener / detail
17. Setup nightly releases for TensorFlowLiteSwift and TensorFlowLiteObjC pods — gardener / detail
18. Disallow setting the forward compatibility horizon in the past. — gardener / detail
19. Less pointer indirection for TFE_OpAttrs, add TFE_OpGetAttrs — gardener / detail
20. Update ops-related pbtxt files. — gardener / detail
21. Add BULID rules for mlir-vulkan-runner — gardener / detail
22. Refresh tf_generated_ops.td with new op definition changes — gardener / detail
23. [tfls.util] Add Category class and conversion from TensorLabel to List<Category>. — gardener / detail
24. [Build cleanup] Remove "tensorflow/core/graph/graph_constructor.h". — gardener / detail
25. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
26. Fix some small issues to enable Kumamon gpu inference: — gardener / detail
27. Add tf.DecodeAndCropJpeg, tf.DecodeGif, tf.DecodeJpeg and tf.DecodePng to TensorFlow MLIR ODS. — gardener / detail
28. Support packed tensor inputs in ProcessFunctionLibraryRuntime. — gardener / detail
29. Convert BFLOAT16 and HALF type tensors to DenseElementsAttr — gardener / detail
30. Remove references to "tf.broadcase_weights" use "tf.broadcast_to". — gardener / detail
31. Update ReplicateToIslandPass to only create a sink island pinning control dependencies instead of forwarding all replica data results. — gardener / detail
32. Fix crash on OpaqueElementsAttr in CreateLiteralFromAttr helper — gardener / detail
33. Constant evaluation of XLA HLO Concatenate — gardener / detail
34. Fix the gradient of x**0 wrt the base (it's now zero instead of NaN) for real dtypes when x is 0 or NaN. — gardener / detail
35. Add AbstractContextInterface::StartStep() and AbstractContextInterface::EndStep() which are used in training models. — gardener / detail
36. For tf.function inputs with nested values, do not require that either zero or all TensorSpecs in the nested value need to have names. — gardener / detail
37. Lower ops with dynamic result type in fallback TF to HLO lowering — gardener / detail
38. Integrate LLVM at https://github.com/llvm/llvm-project/commit/bc23c1d85e4c — gardener / detail
39. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
40. Reference our article where these concepts and computations were introduced, in the hope that the added context and explanations here are helpful to some readers. — gardener / detail
41. Bump open source llvm revision to bc23c1d85e4c4cb60c4b91a1f83466f721014768 — gardener / detail
42. Set logs to a sensible default in case the loop does not run. — gardener / detail
43. Enable some tests : ) — gardener / detail
44. Add test to validate HLO export error on opaque attribute — gardener / detail
45. Add flags to configure whether to fallback to the driver for ptx -> SASS. — gardener / detail
46. Use persistent and scratch buffers in xtensa_hifimini/svdf — gardener / detail
47. Constant folder for TF.Empty — gardener / detail
48. Use persistent buffer in conv and xtensa_hifimini/conv. — gardener / detail
49. Use type inferfence op interface in shape inference pass — gardener / detail
50. Make regularizer docstrings more readable. — gardener / detail
51. Move assign* and scatter* override to DistributedVariable — gardener / detail
52. Internal change — gardener / detail
53. Use persistent buffer in depthwise_conv and xtensa_hifimini/depthwise_conv. — gardener / detail
54. XLA HLO constant folders (add, subtract, multiply) — gardener / detail
55. (Rolling forward after rollback) — gardener / detail
56. Save XPlane by default for TF profiler. — gardener / detail
57. Enable ctl_correctness_test : ) — gardener / detail
58. Support multiple GPUs cross replica reduces in XRT. — gardener / detail
59. Rollback code introduced in #38848 — gardener / detail
60. Remove the extra period in _validate_args (data_adapter.py)'s error message. — gardener / detail
61. [XLA:SPMD] Add API for the experimental mechanism to mix auto and manual partitioning — gardener / detail
62. Deprecate v2 version of group. — gardener / detail
63. Fix a broken multi-gpu test in collective_all_reduce_strategy_test. — gardener / detail
64. Lower TF TruncatedNormalOp using the fallback lowering pass — gardener / detail
65. Internal change — gardener / detail
66. Register Eigen::half kernels for Diag and DiagPart — gardener / detail
67. Deprecate v2 version of group. — gardener / detail
68. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
69. Re-enable keras_stateful_lstm_model_correctness_test as it seems passing now (might have been some env issue with TPUs). — gardener / detail
70. Compute values as shapes where requested by shape function — gardener / detail
71. [XLA:TPU] Move identical ops out of conditional. — gardener / detail
72. [TF:Grappler] Add Quantize/Dequantize ops to IsDefaultLayoutAgnosticOp. — gardener / detail
73. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
74. Fix XLA HLO op import mapping of output_feature_dimension — gardener / detail
75. Add `strict` argument to `resize_tensor_input` in TFLite Python API. — gardener / detail
76. Comment typo fix. — gardener / detail
77. Fix a number of deserialization error messages. While `print(string, obj)` is valid syntax, `ValueError(string, obj)` is not. — gardener / detail
78. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
79. Add `strict` argument to `ResizeInput` in Java API. — gardener / detail
80. Enable __array__ access on EagerTensors (forwarding to _numpy) — gardener / detail
81. Update CVE number to new number issued by MITRE — gardener / detail
82. Extend XRT utils visibility. — gardener / detail
83. Implement DistributedVariable.numpy() — gardener / detail
84. Use the calculated activation range in depthwise_conv — gardener / detail
85. Add flags to configure whether to fallback to the driver for ptx -> SASS. — gardener / detail
86. Update GraphDef version to 392. — gardener / detail
87. compat: Update forward compatibility horizon to 2020-05-05 — gardener / detail
88. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#674 (May 4, 2020, 11:47:00 AM)

1. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
2. Bump open source llvm revision to 57818885be5160380e29e9c2e915b37d5b11ade9 — gardener / detail
3. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
4. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
5. Fix minor errors in micro speech example — gardener / detail
6. Minor updates to micro_speech example. Fix links and images. — gardener / detail
7. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
8. Create ConvertXPlaneToMemoryProfile function for OSS TF memory profiler. — gardener / detail
9. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
10. compat: Update forward compatibility horizon to 2020-05-04 — gardener / detail
11. Update GraphDef version to 391. — gardener / detail
12. Integrate LLVM at https://github.com/llvm/llvm-project/commit/6fe20a44fd3f — gardener / detail
13. Bump open source llvm revision to 6fe20a44fd3fc95198b3b38dc1080266be4ef004 — gardener / detail

#673 (May 3, 2020, 11:47:00 AM)

1. Internal cleanup: let each converter only perform the analysis it needs. Since the bottleneck is inside the conversion process, this is expected to speed it up noticeably. — gardener / detail
2. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
3. [XLA:CPU] Pass negative and positive cpu flags to LLVM — gardener / detail
4. [PJRT] Increase minimum thread stack size for HostStream threads to 8MiB, which was the previous default on Linux. — gardener / detail
5. Log pass timing in a file when the bridge logging is enabled — gardener / detail
6. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
7. Make ConcreteFunction.pretty_printed_signature deterministic. — gardener / detail
8. Fail faster in fully_connected_test — gardener / detail
9. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
10. Revert "Create ConvertXPlaneToMemoryProfile function for OSS TF memory profiler." — gardener / detail
11. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
12. [XLA:coverage] Remove test that times out on coverage //third_party/tensorflow/compiler/xla/tests:convolution_test_viperfish_lite_iss — gardener / detail
13. Update GraphDef version to 390. — gardener / detail
14. compat: Update forward compatibility horizon to 2020-05-03 — gardener / detail
15. Integrate LLVM at https://github.com/llvm/llvm-project/commit/57818885be51 — gardener / detail
16. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#672 (May 2, 2020, 11:47:00 AM)

1. Change code examples in tf.strings.format to use TF 2.x — yong.tang.github / detail
2. Add gcs_file_system to be part of the pip wheel build — yong.tang.github / detail
3. Change to use >>> and ... in docstring format, based on review comment — yong.tang.github / detail
4. Remove explicit assert — yong.tang.github / detail
5. Update tensorflow/python/ops/string_ops.py — github / detail
6. Update tensorflow/python/ops/string_ops.py — github / detail
7. fix doctest failure — yong.tang.github / detail
8. fix trivial typo — ishizaki / detail
9. Remove unneeded "```" in docstring as doctest does not use "```" to locate code block (use ">>>" instead) — yong.tang.github / detail
10. Replace with a shorter example, and removes pylint disable — yong.tang.github / detail
11. Adding const qualifier to a few functions inorder to keep up — srinivasan.narayanamoorthy / detail
12. minor fix. — srinivasan.narayanamoorthy / detail
13. Avoid partially creating/updating cluster when some workers fail during update. — gardener / detail
14. Use correct RBE platform for our CPU RBE config. — gardener / detail
15. Refactor rbe_config.bzl to avoid calling platform / python config multiple times. — gardener / detail
16. Switch tensorflow::Env() to use pthread_create() to start threads instead of std::thread. — gardener / detail
17. Update tf.function's experimental_relax_shapes to handle composite tensors appropriately. — gardener / detail
18. Changes TFL BatchMatMul's inputs/options to be identical to TF versions. — gardener / detail
19. Make it harder to call the private RaggedTensor constructor from user code. — gardener / detail
20. Make it harder to call the private RaggedTensor constructor from user code. — gardener / detail
21. Make test for sparse tensor more generic. — gardener / detail
22. [tfdbg2] Fix debugger-generated names for Placeholder and Const ops — gardener / detail
23. Disable quantize_and_dequantize_op_test test on Windows — gardener / detail
24. Support native Raspberry Pi build for XNNPACK delegate — gardener / detail
25. Move tf.keras.experimental.SequenceFeatures to keras package. — gardener / detail
26. Delete outdated ops history. — gardener / detail
27. Split non-Python PJRT classes into their own directory. — gardener / detail
28. Modified convolution reader to support dynamic weights. — gardener / detail
29. Remove tpu specific dependencies for Cast. — gardener / detail
30. Add a bfloat16 registration for MulNoNan — gardener / detail
31. [XLA:coverage] Disable tests timing out on coverage. — gardener / detail
32. Add clarification to tf.clip_by_norm docs — gardener / detail
33. Rolling forward "[ParseSequenceExampleOp] Optimize config construction." — gardener / detail
34. Add the same scale decompose function — gardener / detail
35. Convolution performance improvement for Adreno. — gardener / detail
36. Re-enable keras_dnn_correctness_test as it appears no longer flaky. — gardener / detail
37. Raise a better error message when a DescriptorPool lookup fails because of a missing BUILD dep. — gardener / detail
38. Added support of dynamic weights for ConvBuffer1x1. — gardener / detail
39. Fix a destruction ordering issue with custom devices and the TFE_Context — gardener / detail
40. Update ops-related pbtxt files. — gardener / detail
41. Fix flakiness in c_api_remote_test. — gardener / detail
42. Default values for TfLiteHexagonDelegateOptions — gardener / detail
43. [tfdbg2] Various improvements to DebugDataReader for DebuggerV2 — gardener / detail
44. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
45. Fix test by raising a class level exception during setup. — gardener / detail
46. Use scratch buffer API in reference SVDF. — gardener / detail
47. Added support of size = -1 in Slice attributes. — gardener / detail
48. [tf.data] Moving `cardinality` to core API. — gardener / detail
49. Fix tf.angle for real tensors. For x < 0, the angle is pi, not zero. — gardener / detail
50. [XLA:Python] Specify a 2MiB stack size for host stream threads. — gardener / detail
51. *** Reason for rollback *** — gardener / detail
52. Added internal embedding ops — gardener / detail
53. [tf.data] Switch tf.data functions to default to using multi-device backend. — gardener / detail
54. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
55. Add transformations for TF/TFLite: — gardener / detail
56. [tf.data] Apply single-device multi component function optimization in `RunSync()`. — gardener / detail
57. Integrate LLVM at https://github.com/llvm/llvm-project/commit/9295f356bb30 — gardener / detail
58. [TF:TRT] Check batch size consistency during segmentation for implicit batch — gardener / detail
59. Disable distribute_strategy_test on oss. — gardener / detail
60. *** Reason for rollback *** — gardener / detail
61. Bump open source llvm revision to 9295f356bb30f283219b3fd32542f0096cfb9316 — gardener / detail
62. Add a CPU bfloat16 kernel for MulNoNan — gardener / detail
63. *** Reason for rollback *** — gardener / detail
64. [tf.data] Fix new autotune + fixed buffer size behavior. — gardener / detail
65. [tf.data] Add trace for IteratorGetNextAsOptionalOp::DoCompute — gardener / detail
66. Add the support for IteratorGetNextAsOptionalOp. — gardener / detail
67. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
68. Correct the error message. — gardener / detail
69. Create interfaces for distributed iterators and iterables. These interfaces are private for now. Use them to break the circular dependency with AutoGraph. — gardener / detail
70. [tf.data] Fix new autotune + fixed buffer size behavior. — gardener / detail
71. Create ConvertXPlaneToMemoryProfile function for OSS TF memory profiler. — gardener / detail
72. Add "alwayslink = 1" to the "quantize_training" target. — gardener / detail
73. [tf.data] Make "DeleteIterator" op a `HybridAsyncOpKernel`. — gardener / detail
74. [OpKernelContext] Enable plumbing the executor type through to kernels. — gardener / detail
75. Make sure custom losses + metrics get autographed, even though autograph marks code in core Tensorflow with `DoNotConvert` — gardener / detail
76. Make sure custom losses + metrics get autographed, even though autograph marks code in core Tensorflow with `DoNotConvert` — gardener / detail
77. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
78. Add BoundingBox helpers. — gardener / detail
79. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
80. [XLA] Add Cbrt operation (lowered to pow(1.0/3.0)) — gardener / detail
81. Update GraphDef version to 389. — gardener / detail
82. compat: Update forward compatibility horizon to 2020-05-02 — gardener / detail

#671 (May 1, 2020, 11:47:00 AM)

1. Enable dynamic shape tests for ConvertSqueeze — tfeher / detail
2. Include .h.inc file into the header files of the wheel package — yong.tang.github / detail
3. Improve docstring, and address issues from reviewer — tfeher / detail
4. Make ConvertSqueeze a value parameterized test — tfeher / detail
5. Update iOS related rules — keithbsmiley / detail
6. Add more test value parameters for ConvertSqueeze, improve helper functions — tfeher / detail
7. Fix regex for scope name specification — yong.tang.github / detail
8. Add test case for scope name regex — yong.tang.github / detail
9. Update test case to escape correctly, and add dash in part of the example — yong.tang.github / detail
10. Integrate LLVM at https://github.com/llvm/llvm-project/commit/52eb2f65a7d2 — gardener / detail
11. Bump open source llvm revision to 52eb2f65a7d28bb225ca8a0bc8c4090d324e22d9 — gardener / detail
12. Use the new ruy API for caching constant matrices. — gardener / detail
13. Improve error message for invalid TensorSpecs in func_graph.py — gardener / detail
14. Add absl::Cord support to RAM file system so that internal ops/tests can use cords — gardener / detail
15. Add a tf-lite only python api-reference generator — gardener / detail
16. Replace spaces with underscore when making scope name. — gardener / detail
17. Add an einsum layer. — gardener / detail
18. Fix the gradient of x**0 wrt the base (it's now zero instead of NaN) for real dtypes when x is 0 or NaN. — gardener / detail
19. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
20. [Single-threaded executor] Optimize the handling of `OpKernel::const_tensor()`. — gardener / detail
21. Use persistent buffer in reference fully_connected. — gardener / detail
22. Clearly indicate when to disable optimizations — gardener / detail
23. Clean up dead core. — gardener / detail
24. Don't propagate ref type if it wasn't originally done — gardener / detail
25. Add the scale propagation pass to the pipeline — gardener / detail
26. Initial checkin of SavedModelAPI C++ header-only API. Tensor, TensorHandle and ConreteFunction::Run will be added in a subsequent change. See RFC https://github.com/tensorflow/community/pull/207. — gardener / detail
27. Disable max_pool_2d_test under TSAN — gardener / detail
28. Disable flaky ASAN test — gardener / detail
29. Disable flaky ASAN test — gardener / detail
30. Disable flaky test on TAP — gardener / detail
31. Use persistent buffer in xtensa_hifimini/fully_connected. — gardener / detail
32. [WhileV2] Fix potential stack overflow when kernels execute inline. — gardener / detail
33. Add hardswish op to hexagon delegate for UInt8/Int8 inputs — gardener / detail
34. [tf.data] Moving `cardinality` to core API. — gardener / detail
35. Internal change — gardener / detail
36. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
37. Remove async runs of cluster update with pending nodes tests. — gardener / detail
38. TF_LITE_MAYBE_KERNEL_LOG with a check for non-null context — gardener / detail
39. Enable a disable test as it no longer seems flaky. — gardener / detail
40. Treat "test_xla_gpu" as GPU_TEST in "NamedGPUCombination". — gardener / detail
41. Avoid partially creating/updating cluster when some workers fail during update. — gardener / detail
42. [tf.data service] Support __iter__ for tf.data service datasets. — gardener / detail
43. Re-enable multinomial_test, no longer flaky. — gardener / detail
44. Rolling forward "[Single-threaded executor] Optimize the handling of `OpKernel::const_tensor()`." — gardener / detail
45. Add transformations for TFLite: — gardener / detail
46. [TF:TRT] Disable the failing squeeze op test. — gardener / detail
47. Use persistent buffer in reference svdf. — gardener / detail
48. Add missing bug to disabled test target. — gardener / detail
49. Removing v1 optimizer tests from ctl_correctness_test to speed it up — gardener / detail
50. pfor: add support for vectorizing "While" and "StatelessWhile" ops. — gardener / detail
51. Handle constant inputs in xla-legalize-tf-with-tf2xla pass — gardener / detail
52. Remove reset_test argument from test_util.run_in_graph_and_eager_modes — gardener / detail
53. Update LLVM hash, manually, to fix Windows builds. — gardener / detail
54. Disable OwnedMultiDeviceIterator on TPU Pod. — gardener / detail
55. Add no_oss tag to custom_training_loop_input_test. — gardener / detail
56. Added converter of BHWC tensor to convolution weights (in specific layout). — gardener / detail
57. Add barebones TPUExtractHeadTailOutsideCompilation Pass. — gardener / detail
58. Enable lowering xla_hlo.convert to linalg.generic op. — gardener / detail
59. Allow functions with raw Send & Recv to be inlined — gardener / detail
60. Improve variable naming consistency in data_service_dataset_op — gardener / detail
61. Typo fix: `==` should be `=`. — gardener / detail
62. API to provide custom table level optimizers in TPUEmbedding. — gardener / detail
63. Initial checkin of C++ header-only Tensor API as part of RFC https://github.com/tensorflow/community/pull/207. — gardener / detail
64. Remove Layer._supports_ragged_inputs property. — gardener / detail
65. Enable previously disabled testUnevenDatasetBatchesBetweenGraph — gardener / detail
66. Support all applicable axes inputs to Concat — gardener / detail
67. Partial rollback of "[Single-threaded executor] Optimize the handling of `OpKernel::const_tensor()`." — gardener / detail
68. Refactors MemorySpaceAssignment to make it easier to subclass. — gardener / detail
69. Add transformations for TFLite: — gardener / detail
70. Added support of any weights sizes to ConverterToConvWeights. — gardener / detail
71. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
72. Add basic gpu support in TF-TFRT integration. When compiling with --config=cuda, tfrt will automatically detect the gpu device. — gardener / detail
73. [XLA]Add HLO infrastructure to support dynamic op lowering. — gardener / detail
74. Change Cord iteration spelling under PLATFORM_GOOGLE. — gardener / detail
75. Re-enabling ps test on OSS. It seems to no longer fail. — gardener / detail
76. Rollback of "Avoid partially creating/updating cluster when some workers fail during update." as it breaks on Windows — gardener / detail
77. Add tf.sparse.bincount op. — gardener / detail
78. Typo fix: tsan should have been asan as the test is flaky on ASAN — gardener / detail
79. Added support of dynamic weights to ConvPowerVR. — gardener / detail
80. Fix incompatibilities between DistributedIterator and the corresponding DistributedIteratorSpec. — gardener / detail
81. [ParseSequenceExampleOp] Optimize config construction. — gardener / detail
82. [tf.data] Switch tf.data functions to default to using multi-device backend. — gardener / detail
83. Disable broken Windows test — gardener / detail
84. Disable flaky test on TSAN — gardener / detail
85. Disable broken Windows test — gardener / detail
86. Enable logging by default about running or not the MLIR bridge — gardener / detail
87. Internal change — gardener / detail
88. Disable flaky guitar test — gardener / detail
89. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
90. [TF2XLA] Clarify the error message when XLA JIT is not available — gardener / detail
91. Internal change — gardener / detail
92. Added selector of Convolution with dynamic weights. — gardener / detail
93. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
94. Add `smart_resize` utility to resize images for use in contiguous batches, without — gardener / detail
95. Open source tf-to-corert and move non-OSS code to — gardener / detail
96. Improve comment for Optimize enum. — gardener / detail
97. Disable specific tests for TPU instead of entire test file. — gardener / detail
98. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
99. Update GraphDef version to 388. — gardener / detail
100. compat: Update forward compatibility horizon to 2020-05-01 — gardener / detail
101. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#670 (Apr 30, 2020, 11:47:00 AM)

1. Make BasicInterpreter.ThreeStepAllocate run on big-endian machines — frreiss / detail
2. Fix second instance of the same problem, and rename my_data -> data — frreiss / detail
3. refactoring: 'if' syntax deduplication — rladhkstn8 / detail
4. fix: Resolving incorrect value change error — github / detail
5. Enable TensorFlow Lite for Microcontrollers to build with CMSIS-NN — dan / detail
6. Integrate CMSIS-NN optimized function for max pooling — dan / detail
7. refactoring: early return 'if else' -> 'if' — rladhkstn8 / detail
8. refactoring: deduplication — rladhkstn8 / detail
9. refactoring: early return 'if else' -> 'if' — rladhkstn8 / detail
10. add case when errors is garbage collected — zilinzhu / detail
11. Fix docstring error of MlirPassthroughOp — yong.tang.github / detail
12. fix doc — huang.dexter / detail
13. Ensure `ParseOpData` always has a return value — dan / detail
14. Auto-populate main with the default so that for simple usage, users do not need to supply it. — gardener / detail
15. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
16. Rollback of Rollback ram filesystem again as it still breaks Windows — gardener / detail
17. Fix build failures for python 3.8 — yibairen.byron / detail
18. Replace tp_print to tp_vectorcall_offset for python 3.8 — yibairen.byron / detail
19. Add RBE configuration that allows building against multiple python versions. — gardener / detail
20. Isolate TFLM project file dependencies. — gardener / detail
21. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
22. Add distribution strategy tests for Keras preprocessing layers. — gardener / detail
23. Update note about latest version — gardener / detail
24. [Build cleanup] Update #includes of moved header "graph/graph_constructor.h". — gardener / detail
25. Removing experimental from run_functions_eagerly. Also removed all references for the experimental one from documentation. — gardener / detail
26. [XLA] Lower verbosity of "Unexpectedly high number of iterations" warning. — gardener / detail
27. Integrate LLVM at https://github.com/llvm/llvm-project/commit/f9106e85c424 — gardener / detail
28. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
29. Migrate our clang configuration to the new multi-python setup. — gardener / detail
30. Use strategy default device (if not None) as the destination for strategy.reduce. — gardener / detail
31. Reduce overuse of inline. — gardener / detail
32. Bump open source llvm revision to f9106e85c424756bcd1378d6a00bb582cc2114bb — gardener / detail
33. Some cleanup of reference fully connected. — gardener / detail
34. Run shape inference after LegalizeTF pass. — gardener / detail
35. Expose enable_v2_dtype_behavior as V1 symbol. — gardener / detail
36. Api doc fixit: Add executable example in `tf.keras.layers.Dense` and argument list in `__init__`. — gardener / detail
37. [XLA:Python] Remove the xla_client Backend class. — gardener / detail
38. Internal code change — gardener / detail
39. Remove redundant parenthesis in dataset_ops.py function `map`'s docstring. — gardener / detail
40. NFC: Disable clang format for op whitelist in xla-legalize-tf-with-tf2xla pass — gardener / detail
41. Enclose the set_cacheable in #ifdef TFLITE_WITH_RUY_GEMV. — gardener / detail
42. Don't crash for unranked tensor types. — gardener / detail
43. Use strategy default device (if not None) as the destination for strategy.reduce. — gardener / detail
44. [Build cleanup] Remove reference to moved files from "tensorflow/core/graph/BUILD". — gardener / detail
45. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
46. Internal test change for Android — gardener / detail
47. Move advanced AutoGraph material from tf.function guide into AutoGraph docs. — gardener / detail
48. Set up TFRT OSS dependency in Tensorflow. — gardener / detail
49. Add tracking bug for broken DistStrat test. — gardener / detail
50. Update comment to reference latest update. — gardener / detail
51. Update ruy to get the new cache API. — gardener / detail
52. Update embedding_ops doc: — gardener / detail
53. Lower precisions in autocast_variable_test — gardener / detail
54. [XLA] Dump debugging information from memory_space_assignment. — gardener / detail
55. In TPUExtractOutsideCompilation op, correctly return the outputs from TPU parallel_execute region. — gardener / detail
56. Allow custom backend legalization pass to compiler — gardener / detail
57. Deprecate tf.keras.experimental.terminate_keras_multiprocessing_pools. This is not tested and not used, as far as we can tell. — gardener / detail
58. Deprecate TF_USE_CUDNN env var. — gardener / detail
59. XLA SliceOP folding support — gardener / detail
60. Add FunctionLibraryRuntime::RunSync() methods that avoid the need for creating a callback. — gardener / detail
61. Fix a missing default case statement. — gardener / detail
62. [tf.data service] Refactor data service RPCs into DataServiceClient classes. — gardener / detail
63. Rewrite output state to match the input. — gardener / detail
64. Disable split_v_op test on Windows — gardener / detail
65. Handle cases where the SliceOp is a Noop — gardener / detail
66. Remove the requirement for the loop variable to be defined before the loop when used in `for ... in tf.range`. — gardener / detail
67. Concatenation zero-length operand removal — gardener / detail
68. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
69. Fix TFE_ContextClearCaches for TFRT. — gardener / detail
70. Add distribution strategy tests for normalization. — gardener / detail
71. Tweaked slice op to avoid reinterpret_cast — gardener / detail
72. Fixed StructuredTensor.partition_outer_dimension. The shape — gardener / detail
73. Allow unsigned integers in HLO_DimensionTensor and HLO_StaticShapeTensor — gardener / detail
74. [tf.data] Supporting shuffle + repeat fusion in TF 2 (and with correct `reshuffle_each_iteration` semantics). — gardener / detail
75. Use the same strategy in testEagerDistributionStrategy — gardener / detail
76. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
77. Let Defun helper methods able to recognize TF's new FuncGraph. — gardener / detail
78. Removing `inputs` arg from Layer `add_loss` API. — gardener / detail
79. Update the doc in map() where the need to use python function has a few options instead of two. — gardener / detail
80. In DistributedIteratorSpec.most_specific_compatible_type, combine _element_specs recursively. — gardener / detail
81. Update ops-related pbtxt files. — gardener / detail
82. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
83. Let Defun helper methods able to recognize TF's new FuncGraph. — gardener / detail
84. Fix lint violation to be 3.8 aware — gardener / detail
85. TFLM: Bumping up the kMaxChannels for Conv/DepthwiseConv to 1024 — gardener / detail
86. Simplify, speed up and improve accuracy for xtensa fixedpoint utils. — gardener / detail
87. Fix a bug in the format converter. — gardener / detail
88. Add tf_device.cluster and tf_device.cluster_func to tf_device dialect. — gardener / detail
89. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
90. Store single _initialize_op when loading SavedModel. — gardener / detail
91. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
92. Update GraphDef version to 387. — gardener / detail
93. compat: Update forward compatibility horizon to 2020-04-30 — gardener / detail
94. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#669 (Apr 29, 2020, 11:47:00 AM)

1. _ConstantValue can now see through tf.identity ops — 7938269+ngc92 / detail
2. fix dynamic compare broadcast — oyjmical / detail
3. refactor deplicate code — oyjmical / detail
4. delete blank line — oyjmical / detail
5. test fixes? — 7938269+ngc92 / detail
6. Update snappy library to 1.1.8 — yong.tang.github / detail
7. remapper test fix. — srinivasan.narayanamoorthy / detail
8. [ROCm] Updating ROCm CI scripts to skip //tensorflow/lite/... — deven.desai.amd / detail
9. rename function name to make them consistent — oyjmical / detail
10. Add missing mlir_passthrough_op in pip wheel install — yong.tang.github / detail
11. fix format — oyjmical / detail
12. Fix incorrect usage of os.system(). — buchgr / detail
13. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
14. Enable setting shape of _Arg tensor for shape of resource args — gardener / detail
15. Restructure TF Lite API doc — gardener / detail
16. [XLA:Python] Harmonize the APIs of the pybind11-bound C++ Client abstraction and the Python Backend class. — gardener / detail
17. Integrate LLVM at https://github.com/llvm/llvm-project/commit/ef06016d7339 — gardener / detail
18. [tfls.image] Add TensorImage.fromBitmap shortcut. — gardener / detail
19. Add BUILD files for the java generator. — gardener / detail
20. Update LLVM to ef06016d73390d5b380018cc0d16003b4ed4a35a — gardener / detail
21. Document FULLY_CONNECTED support in XNNPACK delegate — gardener / detail
22. Remove **kwargs from `fit`. It was never validated, which has the unfortunate side effect that misspelled keyword arguments are accepted despite doing nothing (e.g. `val_data`, `call_backs`, `epohcs` or whatever else). — gardener / detail
23. [Build cleanup] Split "core_cpu_impl" into fine-grained targets (4/n). — gardener / detail
24. Move shape inference pass to be before inliner in CreateTFStandardPipeline. — gardener / detail
25. Remove pattern to map xla_hlo.divide to tf.RealDiv. — gardener / detail
26. Simplify Model.run_eagerly logic for eager-only Model class. — gardener / detail
27. Fixing a function name typo, and removing TF_DeleteConcreteFunction. — gardener / detail
28. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
29. Export all symbols for TFLite C++ .dll — gardener / detail
30. Restrict `xla_hlo.constant` to always have same type for value attribute and output. — gardener / detail
31. Splitting TF_LoadSavedModel into two functions, one that takes tags, and one that doesn't. — gardener / detail
32. Update tf.FusedBatchNormV3 legalization to support exponential_avg_factor. — gardener / detail
33. Emit @tf.function retracing warning only in eager execution. — gardener / detail
34. Move metadata_collector under cuda_deps to fix portability issues on cast. — gardener / detail
35. Break diag_op_test into more shards to prevent timeout. — gardener / detail
36. Small cleanups for experimental TFE attribute APIs — gardener / detail
37. [Profiler] Add tf-function stats to be part of op_stats. — gardener / detail
38. Create Keras Optimizer non slot variables inside strategy scope if the optimizer is created inside strategy scope. — gardener / detail
39. Support scalar inputs in Java TFLite API — gardener / detail
40. More allowed mutations on trackable lists without trackable components — gardener / detail
41. Add pass in MLIR to compress sparse tensor. — gardener / detail
42. Add tests for SyncOnReadVariable.scatter* — gardener / detail
43. Register int64 kernels for BroadcastTo — gardener / detail
44. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
45. Remove distribute test libraries from python:no_contrib — gardener / detail
46. Add memory warning for broadcast_to docs. — gardener / detail
47. Branch TransformTensor operation to TransformTensorV2. — gardener / detail
48. Allow CLion to include targets that depend on Python-generated headers. — gardener / detail
49. Update hexagon guide to link to the new hexagon libraries version v 1.17 — gardener / detail
50. Allow Philox to be fed with 2 elements state. — gardener / detail
51. Fix handling of PartitionedCall in ShapeInference pass. Previously, the pass — gardener / detail
52. Make the keras background graph for functional api construction thread-local, so models can be constructed and trained in separate threads simultaneously. — gardener / detail
53. Change PyLocal* names to PjRt*, where PjRt stands for "pretty much — gardener / detail
54. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
55. Update ReplicateToIslandPass to generate sink islands with fetch operand types instead of replicate island result types. — gardener / detail
56. array_ops.meshgrid should correctly handle no/empty args — gardener / detail
57. Refactor SyncOnReadVariable update methods to be consistent with MirroredVariable. — gardener / detail
58. Document and test explicit padding for separable_conv2d. — gardener / detail
59. Update java generator. — gardener / detail
60. Correct typo and better doc. — gardener / detail
61. Refine result type of std.tensor_cast op during shape inference. — gardener / detail
62. Fix flaky loss_scale_optimizer_test. — gardener / detail
63. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
64. [RunHandler] Minor optimizations to environment variable handling. — gardener / detail
65. Move tf.StringFormat from tf_generated_ops.td to tf_ops.td (NFC). — gardener / detail
66. Update description of tf.BroadcastTo in TensorFlow MLIR ODS to reflect description in TensorFlow op registry (NFC). — gardener / detail
67. Enable setting an op resolver for the tflite model reader. — gardener / detail
68. Deprecate Layer `get_updates_for` and `get_losses_for` APIs. — gardener / detail
69. Attempt to fix certain testable docstrings. — gardener / detail
70. Internal change — gardener / detail
71. _Arg node requires _handle_dtypes and _handle_shapes attr when T is DT_RESOURCE. — gardener / detail
72. Created a new class, SchedulerState, to encapsulate all of the scheduler state-related functionality we would like to reuse across different scheduler implementations. Scheduler state-related member functions/variables in VirtualScheduler have been moved to SchedulerState accordingly. — gardener / detail
73. Internal code change — gardener / detail
74. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
75. Upgrade hexagon_nn to v1.17 — gardener / detail
76. Introduce non-blocking component function execution. — gardener / detail
77. Update Eigen to: https://gitlab.com/libeigen/eigen/-/commit/ab773c7e914633ec4a3ee1f7cdea8b168d3bce1a — gardener / detail
78. Avoid pointer indirection in handle, context & op — gardener / detail
79. Update Eigen commit used by TensorFlow to 1e41406c362788057b3adcd9a25b73f43e6e6492. — gardener / detail
80. Test TextVectorization with all distribution strategies. — gardener / detail
81. Small additions to DistributedStrategy's API docs — gardener / detail
82. [Build cleanup] Split "core_cpu_impl" into fine-grained targets (5/5). — gardener / detail
83. Fix the linking issue as reported in https://github.com/tensorflow/tensorflow/pull/38833 — gardener / detail
84. [dataset_test_base.h] Remove *.proto.h includes. — gardener / detail
85. Add a TODO to remove std.tensor_cast special case in tf-shape-inference. — gardener / detail
86. [tf.data] Switch tf.data functions to default to using multi-device backend. — gardener / detail
87. [tfls.util] Add Model.Options class. — gardener / detail
88. Support MediaPipe Convolution2DTransposeBias in XNNPACK delegate — gardener / detail
89. Internal change — gardener / detail
90. Update doc to comply with Swift API change — gardener / detail
91. [tf.data] Switch tf.data functions to default to using multi-device backend. — gardener / detail
92. Roll forward with fix for Windows & PPC. — gardener / detail
93. Add machine types people use as host for TF builds. — gardener / detail
94. Update GraphDef version to 386. — gardener / detail
95. compat: Update forward compatibility horizon to 2020-04-29 — gardener / detail
96. update instruction for cleaning CocoaPods cache. — gardener / detail
97. Only print the ptxas not found if the error code is NOT_FOUND. — gardener / detail

#668 (Apr 28, 2020, 11:47:00 AM)

1. Allow Prefetching on GPU — github / detail
2. Unittest Added to reproduce the bug — contact / detail
3. Improved Unittest — contact / detail
4. iterator.get_next() colocation bug also fixed — contact / detail
5. Additional GPU Prefetch Fix — contact / detail
6. Formatting fix tensorflow/python/data/experimental/kernel_tests/prefetch_to_device_test.py — contact / detail
7. Update prefetching_ops.py — github / detail
8. Update dataset_ops.py — github / detail
9. Update prefetching_ops.py — github / detail
10. Change ops.colocate_with to ops.device — contact / detail
11. Change ops.colocate_with to ops.device in Iterator — contact / detail
12. Change ops.colocate_with to ops.device in Iterator - 2 — contact / detail
13. Unittest fix — contact / detail
14. Unittests modifications as requested — contact / detail
15. Changes Requested — contact / detail
16. Bug Fix — contact / detail
17. Enabling bf16 fusions — mahmoud.abuzaina / detail
18. replace std::round() with TfLiteRound(). — peng.sun / detail
19. Unittest Changes requested — contact / detail
20. Linting Issues Fix — contact / detail
21. Linting Issues Fix — contact / detail
22. substitute regex to match multi-line timeline_label — zilinzhu / detail
23. Update set of -std values supported by nvcc. — gardener / detail
24. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
25. Rollback ram filesystem again as it still breaks Windows — gardener / detail
26. Remove '-std=c++11' from GPU crosstool compiler invocation. — gardener / detail
27. Update StructuredTensor to use merged RowPartitions. — gardener / detail
28. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
29. TF BatchMatMul/V2 -> TFL BatchMatMul when batch size is unknown. — gardener / detail
30. [TF:TRT] Avoid generating segments that won't be accepted for TRTEngineOp. — gardener / detail
31. Verify replica count from AllReduce replica group config — gardener / detail
32. TF BatchMatMul/V2 -> TFL BatchMatMul when batch size is unknown. — gardener / detail
33. [ExpandDimsOp] Micro-optimizations for `tf.expand_dims()`. — gardener / detail
34. Remove warning upon using generator inputs with multiprocessing. — gardener / detail
35. Integrate LLVM at https://github.com/llvm/llvm-project/commit/8c8aae852b5e — gardener / detail
36. Bump open source llvm revision to 8c8aae852b5e60929156054808af941fc0745d46 — gardener / detail
37. Internal change — gardener / detail
38. Only generate xla_hlo.constant where result and attribute types match exactly. — gardener / detail
39. [Build cleanup] Split "core_cpu_impl" into fine-grained targets (3/n). — gardener / detail
40. Fixed Add fusion in OpenCL. — gardener / detail
41. Use DenseStringElementsAttr for transporting string arrays — gardener / detail
42. OSS the memory_profile.proto for TF profiler. — gardener / detail
43. [Grappler] Allow feeding non-zero index tensors to Grappler optimizer — gardener / detail
44. Fix / standardize references in Keras Applications docstrings. — gardener / detail
45. [XLA:Python] API cleanups. — gardener / detail
46. Saturate when converting image between datatypes in adjust_jpeg_quality. — gardener / detail
47. Fixes:GitHub Issue #38640 — gardener / detail
48. [XLA:Python] Make arguments and backend options mandatory arguments to execute_with_python_values. — gardener / detail
49. Change fronend-attribute value format as a string and add missing parsing — gardener / detail
50. Ensure that models with no fused bias into TransposeConv are exported with only three inputs. This is to ensure that these models still run with older builds of the TFlite runtime. — gardener / detail
51. Make few implicit casts explicit to resolve clang warnings — gardener / detail
52. Opensource TensorFlow's Doclava setup. — gardener / detail
53. Use __getattr__ in Optimizer instead of __getattribute__ — gardener / detail
54. Generalize reshape op lowering to cover expansion/collapse of dims. — gardener / detail
55. Improve documentation for the unified C API and perform some minor cleanups — gardener / detail
56. Introduce PackedTensorHandleData to TensorHandle. A PackedTensorHandleData refers to a list of TensorHandles of the same dtype and shape. — gardener / detail
57. [tf.data] Instruct Grappler not to apply tf.data optimizations to the function library. — gardener / detail
58. Fix flakiness in XNNPACK Logistic test — gardener / detail
59. Don't mangle list of types to string attr — gardener / detail
60. Group TF1 profiles using the distributed runtime. — gardener / detail
61. Support steps_per_execution in Model.predict — gardener / detail
62. Ensure distributed_file_utils.remove_temp_dirpath() can be safely called multiple times. — gardener / detail
63. Redo:Adds support for half_pixel_centers & align_corners in TFLite's resizeNN kernel. — gardener / detail
64. Log individual test times to help users track down long-running tests. — gardener / detail
65. Fix RPI CI build failure — gardener / detail
66. MLIR TF2XLA bridge: BatchMatMulV2 lowering for dynamic shapes — gardener / detail
67. Calculate only necessary gradients of captures in while loop — gardener / detail
68. Use __getattr__ in Optimizer instead of __getattribute__ — gardener / detail
69. During import add resource type's subtypes. — gardener / detail
70. tf.keras.layers.Conv: Use most_specific_compatible_shape to determine when to re-build the internal nn_ops.Convolution object. — gardener / detail
71. [mlir] Support xla_hlo.broadcast to linalg.generic conversion — gardener / detail
72. Internal change to build rules — gardener / detail
73. Branch RoIToTransforMatrix operation to RoIToTransforMatrixV2. — gardener / detail
74. Add sanity checks to the following TFL ops — gardener / detail
75. Revise Swift API style — gardener / detail
76. Add optional layout constraint to AllToAll — gardener / detail
77. [tf.data] Fix a bug in input time computation. — gardener / detail
78. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
79. Enable framework_ops_test test for TFRT. — gardener / detail
80. tf.keras.layers.Conv: Use most_specific_compatible_shape to determine when to re-build the internal nn_ops.Convolution object. — gardener / detail
81. Fix the iOS benchmark app build by adding CoreML framework — gardener / detail
82. Enable resnet50 tfrt target.  The actual tests are largely disabled. — gardener / detail
83. [tf.data] Update the processing time functions to solve the stack overflow problem. — gardener / detail
84. TFLite Metal delegate: Memory leak when aggressive wait mode is used has been fixed. — gardener / detail
85. Dataset utilities UX improvements — gardener / detail
86. Add a TFL_CustomOp definition and support in exporter and importer. — gardener / detail
87. Internal change — gardener / detail
88. Add sanity checks to TFLite's MulOp — gardener / detail
89. Display the list of the available platforms in the error message. — gardener / detail
90. compat: Update forward compatibility horizon to 2020-04-28 — gardener / detail
91. Update GraphDef version to 385. — gardener / detail

#667 (Apr 27, 2020, 11:47:00 AM)

1. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
2. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
3. Remove unused pybind for sparsification. — gardener / detail
4. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
5. Introduce a RamFileSystem for TensorFlow — gardener / detail
6. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
7. Moving init_scope from Reduce metric to base class add_weight call. — gardener / detail
8. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
9. Generalize reshape op lowering to cover expansion/collapse of dims. — gardener / detail
10. Generalize reshape op lowering to cover expansion/collapse of dims. — gardener / detail
11. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
12. compat: Update forward compatibility horizon to 2020-04-27 — gardener / detail
13. Update GraphDef version to 384. — gardener / detail
14. Fix a potential overflow bug. — gardener / detail
15. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#666 (Apr 26, 2020, 11:47:00 AM)

1. Make tf_gpu_cc_test always execute the test in CPU only build, and fixed a bug exposed in c_api_test by this change. — ag.ramesh / detail
2. Install multiple python versions simultaneously in our remote build docker. — gardener / detail
3. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
4. Register ram filesystem on Windows too — gardener / detail
5. Refactor the C Unified API TF_AbstractTensor implementation the same way the ExecutionContext is exposed — gardener / detail
6. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
7. Refactor unified C API: extract the struct AbstractOp in the private header — gardener / detail
8. Remove TF_ExecutionContextOptions from the unified C API — gardener / detail
9. Refactor TF_OutputList in the C unified API to align with the other C++ type — gardener / detail
10. Refactor AbstractionFunction implementation to make it abstract on the C++ side as well — gardener / detail
11. Refactor the unified C API: split the eager/graph implementation in different files — gardener / detail
12. Add CompositeTensor support to distributed iterators. — gardener / detail
13. Modernize tf.gradients docstring. — gardener / detail
14. [tf.data] Modifying auto-sharding to remove unsupported `assert_cardinality` transformations (instead of triggering an error). — gardener / detail
15. Re-enable TSAN for remote_cluster_test. — gardener / detail
16. [tf.data] Make `AnonymousSeedGeneratorHandleOp` thread-safe. — gardener / detail
17. Use internal graph indices as delegate indices. — gardener / detail
18. Use global empty tuple/dict. — gardener / detail
19. Stops special-casing first argument in Keras's internal Node representation. This makes the special-casing-behavior in __call__ more explicit, and make it easier to reduce how much special casing of the first argument Keras does moving forward. — gardener / detail
20. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
21. Introduce a RamFileSystem for TensorFlow — gardener / detail
22. [tf.sparse_segment_*()] Grappler optimization to avoid casts for indices/segment_ids. — gardener / detail
23. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
24. [XLA] Turn constant dynamic slices into slices. — gardener / detail
25. Update match dtype logic in LossesContainer. — gardener / detail
26. [tf.data service] Make server_lib visible. — gardener / detail
27. Clean up expired forward compatibility checks. — gardener / detail
28. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
29. Register MLIR Context and PassManager option for tf_tfl_translate command line parsing — gardener / detail
30. Disable node_test on mac. — gardener / detail
31. [tf.data] Warn users that the `tf.config.experimental_run_functions_eagerly` option does not apply to tf.data functions. — gardener / detail
32. Update GraphDef version to 383. — gardener / detail
33. compat: Update forward compatibility horizon to 2020-04-26 — gardener / detail
34. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#665 (Apr 25, 2020, 11:47:00 AM)

1. Update documentation to switch from deprecated tf.extract_image_patches to tf.image.extract_patches — yong.tang.github / detail
2. build fix. — srinivasan.narayanamoorthy / detail
3. Revert skipping part of saved_model_test.test_mnist_cnn. — gardener / detail
4. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
5. [XLA:Python] Lower minimum alignment requirement for zero-copy buffer passing. — gardener / detail
6. Fix XLA Igamma(c) when x >> a. — gardener / detail
7. Created a new class, SchedulerState, to encapsulate all of the scheduler state-related functionality we would like to reuse across different scheduler implementations. Scheduler state-related member functions/variables in VirtualScheduler have been moved to SchedulerState accordingly. — gardener / detail
8. Rename shared_device_buffer to tracked_device_buffer to reflect the — gardener / detail
9. Don't pass invalid args to tp_new, it triggers an assertion in Python 3.7+. — gardener / detail
10. Flip shape inference importer flag default — gardener / detail
11. Add auto generated TensorFlow op ApproximateEqual — gardener / detail
12. Adds support for half_pixel_centers & align_corners in TFLite's resizeNN kernel. — gardener / detail
13. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
14. Fix convolution_2d_transpose_bias without bias — gardener / detail
15. [XLA:Python] Remove some Python 2 compatibility code. — gardener / detail
16. Roll forward with fix for tf-serving. — gardener / detail
17. Properly handle negative shape dimensions from improper saved models. — gardener / detail
18. [XLA:Python] Delete xla_client.{Computation,ComputationBuilder}. — gardener / detail
19. Convert f16 and bf16 element type constants to/from XLA literal — gardener / detail
20. Add folder for BroadcastInDimOp — gardener / detail
21. Make DerivedXLineBuilder public. — gardener / detail
22. Implement `InferShapedTypeOpInterface` interface on all CHLO broadcastable binary elementwise ops. — gardener / detail
23. Implement a custom MLIR attribute in TF dialect for shape attributes. — gardener / detail
24. [Grappler] Disable node-destructing rewrites if node is in the preserve set — gardener / detail
25. Add broadcasting binary op patterns for Complex and Compare. — gardener / detail
26. handle the case where cupti subscribe return CUPTI_ERROR_MAX_LIMIT_REACHED. — gardener / detail
27. Adding tf_copts() to SavedModel libraries exporting public C symbols. Missing this will cause Windows to add a dllimport attribute instead of the dllexport attribute. — gardener / detail
28. Integrate LLVM at https://github.com/llvm/llvm-project/commit/9774c34a84e3 — gardener / detail
29. [tf.data service] Support non-localhost worker address. — gardener / detail
30. Plumb inference enabling flag to inner importer — gardener / detail
31. Bump open source llvm revision to 9774c34a84e336968f276591c200bd955631c5be — gardener / detail
32. Implement test pass for checking behavior of InferShapedTypeOpInterface and add tests for CHLO. — gardener / detail
33. Update tf.MatrixBandPart, tf.ResourceApplyAdagradV2 and tf.ResourceApplyMomentum summaries and descriptions to match between TensorFlow op definitions and TensorFlow MLIR ODS. — gardener / detail
34. [Grappler] Add arithmetic optimizer stage for optimizing `tf.nn.embedding_lookup_sparse()`. — gardener / detail
35. Prevent loading saved models where constant nodes have no tensor value. — gardener / detail
36. Create Keras Optimizer non slot variables inside strategy scope if the optimizer is created inside strategy scope. — gardener / detail
37. Remove the necessity of acquiring context_update_mu_ in cluster update. — gardener / detail
38. [XLA] Turn constant dynamic slices into slices. — gardener / detail
39. Materialize derived shapes as TF_ShapeAttrs. — gardener / detail
40. Don't use version 3 for TransposeConv if the optional bias isn't specified. — gardener / detail
41. Lower xla_hlo.convert to tf.Cast op. — gardener / detail
42. Fix resize check for Java inputs — gardener / detail
43. Non-functional change in preparation for introducing non-blocking function execution. — gardener / detail
44. Flip behavior for SavedModel importer to not use shape inference on import — gardener / detail
45. [Grappler] Fix a bug in folding multiply into add arithmetic optimization — gardener / detail
46. Fix minimal logging build for macos — gardener / detail
47. Refactor the derived_timeline files. — gardener / detail
48. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
49. [XLA] Change backend_config from string to bytes — gardener / detail
50. [XLA] Turn constant dynamic slices into slices. — gardener / detail
51. Add python wrapper to modify model interface for full integer quantization — gardener / detail
52. Introduce unsigned integer types for HLO dialect — gardener / detail
53. Flip behavior for SavedModel importer to not use shape inference on import — gardener / detail
54. RankOp folder for RankedTensorType — gardener / detail
55. Create Keras Optimizer non slot variables inside strategy scope if the optimizer is created inside strategy scope. — gardener / detail
56. Improve testing of the stack operation. — gardener / detail
57. Replace the MACRO in sub with a templated function. — gardener / detail
58. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
59. Fix minimal logging build for macos — gardener / detail
60. Integrate LLVM at https://github.com/llvm/llvm-project/commit/79702dd349f3 — gardener / detail
61. Refactor SyncOnReadVariable update methods to be consistent with MirroredVariable. — gardener / detail
62. Internal change — gardener / detail
63. Improve tf.io.gfile.glob docs. — gardener / detail
64. Refactor the unified C API implementation in a private header — gardener / detail
65. Disabling tensorflow/core/profiler/internal/gpu:device_tracer_test_gpu on TAP. — gardener / detail
66. Add notsan to //third_party/tensorflow/lite/delegates/xnnpack:logistic_test. — gardener / detail
67. [Grappler] Do not rewrite nodes in a feed set in arithmetic and function optimizers — gardener / detail
68. Bump open source llvm revision to 79702dd349f31c0c67bf35f36435fdc843fcd052 — gardener / detail
69. Raise UnImplemented ParameterServerStrategy w/ custom training in TF2 and w/ Keras — gardener / detail
70. Introduce a RamFileSystem for TensorFlow — gardener / detail
71. compat: Update forward compatibility horizon to 2020-04-25 — gardener / detail
72. Update GraphDef version to 382. — gardener / detail
73. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#664 (Apr 24, 2020, 11:47:00 AM)

1. Don't build the RTOS component of Mbed OS in order to save flash/RAM — jens.elofsson / detail
2. Add the micro_recognition example. — jens.elofsson / detail
3. Add micro_recognition testcase. — jens.elofsson / detail
4. Add a README file for the micro_recognition example. — jens.elofsson / detail
5. Fix review comments. — jens.elofsson / detail
6. Fix review comments. — jens.elofsson / detail
7. Remove the model from the repo. — jens.elofsson / detail
8. Download the image recognition model when building. — jens.elofsson / detail
9. Adapt to the move of micro. — jens.elofsson / detail
10. Micro: Remove IMAGE_ARRAY define. — jens.elofsson / detail
11. Move image_recognition to examples/experimental/. — jens.elofsson / detail
12. Add wildcard for tensorflow/lite/micro/examples/experimental/-folder. — jens.elofsson / detail
13. add different pattern to show op in timeline — zilinzhu / detail
14. fix bug — zilinzhu / detail
15. Rename disco/ to stm32f746_discovery/. — jens.elofsson / detail
16. Replace Report() with TF_LITE_REPORT_ERROR-macro. — jens.elofsson / detail
17. add comments and fix typos — zilinzhu / detail
18. correct lint warnings — zilinzhu / detail
19. add check for 0 funcs and raise error — vasanirushabh24 / detail
20. fix typo — vasanirushabh24 / detail
21. add test and fix docstring — vasanirushabh24 / detail
22. Consistent capitalization of Python in array_ops.py — github / detail
23. Update CI build scripts — biagio.hkr / detail
24. [ROCm] Updating Dockerfile.rocm / CI scripts to use ROCm 3.3 — deven.desai.amd / detail
25. Add builtin len() support of tf.data.Dataset for autograph — yong.tang.github / detail
26. Add test case for builtin len() support of tf.data.Dataset in autograph — yong.tang.github / detail
27. Use cardinality to process the infinite case, as was suggested by review comment — yong.tang.github / detail
28. Fallback to reduce to process the unknown cardinality case — yong.tang.github / detail
29. Add test case for unknown and infinite cadinality of dataset — yong.tang.github / detail
30. Pylint fix — yong.tang.github / detail
31. Throw out error for cardinality == UNKNOWN as well — yong.tang.github / detail
32. Update PR based on review comment, for using identity and using tf.function for graph mode — yong.tang.github / detail
33. [ROCm] Fix for a build failure on the ROCm platform - 200323 - — deven.desai.amd / detail
34. Disabling 3 subtests within the //tensorflow/core/grappler/optimizers:constant_folding_test — deven.desai.amd / detail
35. Update test with self.evaluate(), as was suggested in review — yong.tang.github / detail
36. update docstring — github / detail
37. [MLIR][XLA] Add HLO to LHLO legalisation patterns for complex, imag, and real operations. — gardener / detail
38. Bump open source llvm revision to 35cf2f42dda4d708741e06570b2dbe91cec4dc41 — gardener / detail
39. Use splat constructor while create constant in legalization of tf.FusedBatchNormV3 op. — gardener / detail
40. [XLA:GPU] Don't horizontally fuse fusions with DUS roots — gardener / detail
41. [tf.data] This CL changes how the shuffle seed generator is managed, making it possible for the shuffle dataset to support both a) sharing of the seed generator across iterators and b) serialization. As a consequence, this CL enables reshuffling across iterations for tf.distribute and tf.data service use cases (which require both sharing of the seed generator across iterators and serialization support). — gardener / detail
42. Minor cleanup of dead code. — gardener / detail
43. Update ops-related pbtxt files. — gardener / detail
44. Adds support for half_pixel_centers & align_corners in TFLite's resizeNN kernel. — gardener / detail
45. Fix typo "arguements" in Keras base_layer.py — gardener / detail
46. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
47. Add distribution strategy tests. — gardener / detail
48. Export ABSL seed symbol in linker script — gardener / detail
49. Use internal graph indices as delegate indices. — gardener / detail
50. Switch from unordered_map to absl::flat_hash_map — gardener / detail
51. Improve testing of tf.bool in tf.unstack op. — gardener / detail
52. Add chlo -> hlo lowerings for dynamic broadcasts. — gardener / detail
53. Add missing symbols to symbols_pybind. — gardener / detail
54. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
55. Export text_dataset_from_directory to the public API. — gardener / detail
56. [tf.data service] Make master the source of truth for jobs finishing. — gardener / detail
57. Update sequential model to not rely on self._is_graph_network. — gardener / detail
58. Windows toolchains (e.g. MSVC 2017) do not provide C11 ::aligned_alloc — gardener / detail
59. Error out from ScopedAllocatorOptimizer when input shape is not fully known. — gardener / detail
60. [XLA] Move assignment exporting and graph coloring until after the graph is simplified. — gardener / detail
61. Use toco_pybind to invoke model sparsification logic for pip size purpose. — gardener / detail
62. [tf.sparse.segment_*()] Add missing kernel registrations for DT_INT64 indices. — gardener / detail
63. [tf.data] Remove invalid test. — gardener / detail
64. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
65. Nanosecond latency measurement for OVIC. — gardener / detail
66. Ensure TFLite shared libs have resolved symbols — gardener / detail
67. Skip part of saved_model_test.test_mnist_cnn due to failure on guitar. — gardener / detail
68. Avoid re-creating cluster FLR and process FLR when updating worker context. — gardener / detail
69. Update ops-related pbtxt files. — gardener / detail
70. Add serialization backwards compatibility tests — gardener / detail
71. Change visibility rules. — gardener / detail
72. [XLA:Python] Add default arguments to some collective ops. — gardener / detail
73. Use --no-cache-dir more often to save Docker image space — gardener / detail
74. Make shape inference on import optional — gardener / detail
75. Add CompositeTensor support to distributed iterators. — gardener / detail
76. Tensor Tracer: — gardener / detail
77. [XLA] Fix alternate memory reporting. — gardener / detail
78. [grappler] Remove dead code during tensor constant compression. — gardener / detail
79. Make chlo_legalize_to_hlo_broadcasts.mlir actually dynamic. — gardener / detail
80. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
81. Add a SavedModelAPI factory function to AbstractContextInterface, which allows a runtime-agnostic SavedModelAPI implementation.  See RFC https://github.com/tensorflow/community/pull/207. — gardener / detail
82. Add partial ExtractOutsideCompilationPass for extracting CPU computation from TPU cluster. — gardener / detail
83. Disable (ResizeNearestNeighborReference, Test2x2To3x3_AlignCorners) test in resize_nearest_neighbor_test.cc for OSS breakage. — gardener / detail
84. Internal change — gardener / detail
85. Add tfjs optimization pass with the first optimization as transforming tf.keras prelu op sets into tfjs.Prelu op. — gardener / detail
86. Rename SharedDeviceBuffer to TrackedDeviceBuffer since that better describes its purpose now. Also rename BufferDefinitionEvent to BufferSequencingEvent since the class is used for tracking reads as well as writes now. A followup commit will rename the files to reflect the new class name. — gardener / detail
87. MLIR conversion for ScatterND — gardener / detail
88. Add -fno-threadsafe-statics to bluepill build. — gardener / detail
89. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
90. Fully Connected operator in XNNPACK delegate — gardener / detail
91. Fix the bug that flatbuffer importer doesn't import dynamic shape tensor correctly. — gardener / detail
92. Created a new class, SchedulerState, to encapsulate all of the scheduler state-related functionality we would like to reuse across different scheduler implementations. Scheduler state-related member functions/variables in VirtualScheduler have been moved to SchedulerState accordingly. — gardener / detail
93. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
94. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
95. Update GraphDef version to 381. — gardener / detail
96. compat: Update forward compatibility horizon to 2020-04-24 — gardener / detail
97. [XLA:CPU] Reconsider alignment heuristics — gardener / detail
98. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail

#663 (Apr 23, 2020, 11:47:00 AM)

1. This PR is one of steps to extend 8-bit quantization to support symmetric 16-bit activations. — elena.zhelezina / detail
2. Addressed reviewer comments. — elena.zhelezina / detail
3. Tidy up for CONV2D: input/output offset is zero in case of 16-bit activations. — elena.zhelezina / detail
4. When activation is relu, we need to clamp the values. — elena.zhelezina / detail
5. Fix for sanity check with buildifier. — elena.zhelezina / detail
6. Fix for warning: unused variable. — elena.zhelezina / detail
7. Fix for the warning. — elena.zhelezina / detail
8. Updated test and fixed a build error. — elena.zhelezina / detail
9. Boiler plate code for dnnl threadpool and mkl_conv_ops.cc — srinivasan.narayanamoorthy / detail
10. Fix for fsanitize=undefined. — elena.zhelezina / detail
11. review changes. — srinivasan.narayanamoorthy / detail
12. minor cleanup. — srinivasan.narayanamoorthy / detail
13. Upgrading DNNL(previously known as MKL DNN) to v1.4 — ag.ramesh / detail
14. [TFLite] Introduce ruy_profiler to makefile — yu65789 / detail
15. [Intel MKL] Fixes an issue with tensorflow bfloat16 scope that creates node names with '//', which causes unknown node name problem when run inference from a trained checkpoint — xiaoming.cui / detail
16. Adding TF_TRT_BLACKLIST_OP — contact / detail
17. Changing TF_TRT_BLACKLIST_OP to TF_TRT_BLACKLIST_OPS — contact / detail
18. Cleaning Leftovers — contact / detail
19. adds build capability under msys on win64 host; uses unzip for unpacking zip files, corrects emulation of --strip-components=1 with unzip — manuel.loew.infineon / detail
20. Env Var name change — contact / detail
21. sorting header files. — srinivasan.narayanamoorthy / detail
22. [XLA] Add support for bf16 map() to HloEvaluator — gardener / detail
23. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
24. Removed redundant empty spaces in order to have proper links. — gardener / detail
25. [MLIR] Disable SelectAndScatterOp tests temporarily. — gardener / detail
26. Match type of unused outputs of `tf.FusedBatchNormV3` during legalization. — gardener / detail
27. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
28. [tf2tensorrt] Support Conv2DBackpropInput with input_sizes of size 2. — gardener / detail
29. Simplify shape code. — gardener / detail
30. Depend on a specific version of google-api-python-client. — gardener / detail
31. Add rank check to Sub op delegation to NNAPI — gardener / detail
32. Track lambda functions separately in the CFG, consistent with normal function definitions. — gardener / detail
33. [XLA][MLIR] Add complex unary elementwise op definitions to LHLO dialect. — gardener / detail
34. Rolling forward "[tf.unique] Optimize the hash table implementation in `UniqueOp::Compute()`." — gardener / detail
35. Rolling forward "[Build cleanup] Split "core_cpu_impl" into fine-grained targets (1/n)." — gardener / detail
36. [MLIR][XLA] Add complex, imag, and real to lhlo/hlo emitters. — gardener / detail
37. Add Python bindings and some tests for the parallel device — gardener / detail
38. Legalize tf.Round to xla_hlo.round — gardener / detail
39. Fix a wrong header and tweaked the wording a little bit. — gardener / detail
40. Add file identifier and model version to python TFLite FlatBuffers — gardener / detail
41. Create `text_dataset_from_directory` utility and refactor shared code between `text_dataset_from_directory` and `image_dataset_from_directory`. — gardener / detail
42. Make sure context_view_id_ is always initialized. — gardener / detail
43. Internal visibility change. — gardener / detail
44. Internal change — gardener / detail
45. Change `log` to `log1p` in Random Gamma sampler. — gardener / detail
46. Add an option to choose the I/O Device for saving and loading models. — gardener / detail
47. Use ruy's newly curated default set of paths. — gardener / detail
48. Update tf.InfeedDequeueTuple -> xla_hlo.infeed legalization to insert a default sharding (device 0) to account for token result. — gardener / detail
49. rollback of change — gardener / detail
50. add tfstreamz metric capture support. — gardener / detail
51. Add type op interface to tf.const — gardener / detail
52. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
53. PR #34903: TransposeConv with Bias — gardener / detail
54. Move TensorLike under types.internal and rename it NativeObject to distinguish from types which actually behave like tensors. — gardener / detail
55. Fix error and reenable input_lib_test. — gardener / detail
56. Improves tf-numpy's `clip` to pass testClipStaticBounds, and adds uint8 kernels to tf.minimum and tf.maximum. — gardener / detail
57. Add no_gpu tag on tensorflow/python:timeline_test_gpu. — gardener / detail
58. Disabling remote_mirrored_strategy_eager_test in OSS. — gardener / detail
59. Make tensor with dims product equal to 1 be considered as a scalar. — gardener / detail
60. Adapt TFLite to new ruy::Context API. — gardener / detail
61. Exposes should_record_summaries() and optimizes for static record_if values. — gardener / detail
62. Change incorrect padding comment. — gardener / detail
63. Update ops-related pbtxt files. — gardener / detail
64. [XLA][TF2XLA] Support tensor list with dynamic dimension. — gardener / detail
65. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
66. [XLA] Add a RootInstructionSinker pass. — gardener / detail
67. Ensure TFLite shared libs have resolved symbols — gardener / detail
68. [tf.data service] Expose number of master tasks through test server_lib. — gardener / detail
69. Update docstring for tf.group and add note to tf.control_dependencies. — gardener / detail
70. MSVC compatibility in TFLite interpreter test — gardener / detail
71. Add parameter to BuiltinDataAllocator::Allocate for requesting alignment. — gardener / detail
72. Only check Callback batch timings and throw warning once per method. — gardener / detail
73. [Build cleanup] Split "core_cpu_impl" into fine-grained targets (2/n). — gardener / detail
74. [XLA:Python] Make xla_client_test.py parametric in its backend. — gardener / detail
75. Internal change — gardener / detail
76. Minor cleanup for the network inits. — gardener / detail
77. Update TensorFlow MLIR ODS for tf.Maximum and tf.Minimum to support uint8. — gardener / detail
78. Add a launch id field in run options and hlo module config. — gardener / detail
79. Partially add derived attribute materialize functions — gardener / detail
80. Make multiple changes to enable running an inference model on TFRT: — gardener / detail
81. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
82. Fix deps error for evaluation_stages python library — gardener / detail
83. open source task API c code — gardener / detail
84. Support non-constant begin and end inputs to tf.StridedSlice in TF to XLA HLO lowering. — gardener / detail
85. fix the typo — gardener / detail
86. serialize tensorflow::monitoring::Percentiles to xplane. — gardener / detail
87. Better, actionable error message when Model.fit, evaluate, predict, or on_batch — gardener / detail
88. Add memory fragmentation instrumentation for BFCAllocator. — gardener / detail
89. Fix CFG to correctly account for lambda functions that appear in while and for loops. — gardener / detail
90. Write the control flow lowering in native Python and let the transformation rewrite them in functional form. This is a no-op change that simplifies the intermediate code. — gardener / detail
91. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
92. Have XLA CPU emit TraceMe calls by default in JIT mode (roll forward) — gardener / detail
93. [TF2XLA] More generic check for a TensorList crossing the TF2XLA boundary — gardener / detail
94. Add a pass that converts MLIR HLO to MLIR LHLO by using the XLA pipeline — gardener / detail
95. Add a note to `model_to_estimator` docs about weighted metrics. — gardener / detail
96. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
97. Fix typo in mirrored_run.py. — gardener / detail
98. Remove obsolete mlir-level ophint-extraction logic. — gardener / detail
99. [tf.data service] Refactor master data structures to be classes instead of structs. — gardener / detail
100. cast down i64 to i32 for begin/size in slice op as NNAPI does not support i64 begin/size, this should be safe for other targets. — gardener / detail
101. Move hash table op kernels out of experimental directory — gardener / detail
102. Split MlirCompiler into two classes. — gardener / detail
103. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
104. [XLA][MLIR] Enable tests for SelectAndScatterOp lowering. — gardener / detail
105. compat: Update forward compatibility horizon to 2020-04-23 — gardener / detail
106. Update GraphDef version to 380. — gardener / detail
107. Remove no longer used compute capability list macro in C++ header. — gardener / detail
108. Roll forward with fix for tf-serving. — gardener / detail
109. Go: Update generated wrapper functions for TensorFlow ops. — gardener / detail
110. Integrate LLVM at https://github.com/llvm/llvm-project/commit/35cf2f42dda4 — gardener / detail