| Commit message (Collapse) | Author | Age |
|
|
|
| |
Change: 139349604
|
|
|
|
|
|
| |
LoadSessionBundleFromPathUsingRunOptions() fails.
Change: 139348875
|
|
|
|
|
|
|
| |
difference in shapes between the Insert and Import functions was causing issues with a vector key and scalar value input.
Fixed by splitting the LookupInterface CheckKeysAndValueTensors method into one for Insert and the other for Import.
Change: 139346138
|
|
|
|
| |
Change: 139345066
|
|
|
|
| |
Change: 139339828
|
|
|
|
| |
Change: 139328977
|
|
|
|
| |
Change: 139325873
|
|
|
|
| |
Change: 139324859
|
|
|
|
|
|
| |
sparse_values_to_indices op.
Change: 139320375
|
|
|
|
|
|
|
|
|
| |
This makes JPEG go 2x faster on x86_64 (k8), arm7, and arm8. On all
other CPU targets, e.g. x86, JPEG performance should be the same as it
was before.
Fixes #4807
Change: 139295768
|
|
|
|
|
|
| |
This fixes an issue where nodes in a while loop were accumulating a very
large amount of stats, exceeding the protocol buffer limit.
Change: 139277518
|
|
|
|
| |
Change: 139271215
|
|
|
|
| |
Change: 139267342
|
|
|
|
|
|
|
|
|
| |
Intended to make the use of TF_STRING valued tensors in other language bindings
easier.
No added unittests since the existing TestEncodeDecodeStrings test case covers
the changes.
Change: 139263312
|
|
|
|
| |
Change: 139259251
|
|
|
|
| |
Change: 139254648
|
|
|
|
| |
Change: 139254613
|
|
|
|
|
|
|
| |
128 threads.
This increases the parallelism in this kernel, yielding a nice speedup.
Change: 139250945
|
|
|
|
|
| |
change does a 1st order Taylor expansion to solve for small values. Also add a new assert, assert_bijective_and_finite, and corresponding tests for some Bijectors.
Change: 139250783
|
|
|
|
| |
Change: 139247755
|
|
|
|
| |
Change: 139247118
|
|
|
|
| |
Change: 139246371
|
|
|
|
| |
Change: 139240711
|
|
|
|
| |
Change: 139239348
|
|
|
|
|
|
| |
semantic breaking change.
Change: 139237086
|
|
|
|
|
| |
The string piece holds a reference to the internal buffer_ when this class gets destructed the buffer gets destructed but the stringpiece can still be used outside (it is an input parameter). The fix is to first copy into scratch (which is an input parameter) and then put the scratch in the stringpiece (result).
Change: 139236445
|
|
|
|
|
| |
(breaking change)
Change: 139235557
|
|
|
|
|
|
|
| |
(selected, neighbor, hover). Still black + white pending design.
Refactor adapter to use new method styleStringFromPackedRgba, which does the
array indexing and string construction.
Change: 139235290
|
|
|
|
| |
Change: 139229687
|
|
|
|
|
|
|
|
|
| |
values to rerun the shape inference function. This removes the need to pass
input_tensors_needed and input_tensor_as_shapes_needed.
Remove calls to call_cpp_shape_fn that pass input_tensors_needed or
inptu_tensor_as_shapes_needed.
Change: 139228180
|
|
|
|
| |
Change: 139224573
|
|
|
|
|
|
| |
break device assignments set by users.
Change: 139224359
|
|
|
|
| |
Change: 139223923
|
|
|
|
|
| |
the label coloring for selected / hover / neighbor points, coming soon.
Change: 139223883
|
|
|
|
|
| |
Caught by a MemorySanitizer test failure.
Change: 139222793
|
|
|
|
|
|
| |
names than the corresponding numpy function
Change: 139221107
|
|
|
|
|
| |
of a numpy array.
Change: 139220642
|
|
|
|
| |
Change: 139218176
|
|
|
|
| |
Change: 139217997
|
|
|
|
|
| |
in preparation for gRPC debug ops
Change: 139216871
|
|
|
|
| |
Change: 139215742
|
|
|
|
|
|
| |
Also Implemented ReverseV2's gradient in C++
Up the version to indicate the new op is ready to be used.
Change: 139215566
|
|
|
|
| |
Change: 139210204
|
|
|
|
|
|
|
| |
The choco-installed version of SWIG did not include the fix to
https://github.com/swig/swig/issues/567, which caused some Python
tests to fail on Windows.
Change: 139206577
|
|
|
|
| |
Change: 139202892
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Prior to this change, TF_*Run, TF_SetAttrTensor and TF_SetAttrTensorList
took ownership of the TF_Tensor*s of the feeds. This can make performance
client languages bothersome when the same Tensor is repeatedly fed into
multiple session executions as the memory for the feed tensor would need
to be re-allocated and filled in every time.
With this change, these functions no longer take ownership of the
TF_Tensor*.
The changes to the Go API implementation reflect the claimed benefits.
(Another step towards #10)
Change: 139169388
|
|
|
|
| |
Change: 139163917
|
|
|
|
| |
Change: 139161477
|
|
|
|
| |
Change: 139151696
|
|
|
|
| |
Change: 139149048
|