| Commit message (Collapse) | Author | Age |
|
|
|
| |
PiperOrigin-RevId: 211082479
|
|
|
|
|
|
| |
They don't make sense in the open source repository.
PiperOrigin-RevId: 183140889
|
|
|
|
|
|
|
|
| |
The new op is a fused implementation of the existing
`tf.parse_single_example()`, which is more efficient when parsing a
single Example at a time.
PiperOrigin-RevId: 179768512
|
|
|
|
| |
PiperOrigin-RevId: 165220185
|
|
|
|
|
|
|
| |
c++ op.
Full python support (including more comprehensive documentation) coming soon.
Change: 146852707
|
|
|
|
|
|
| |
This is functionally a no-op and should have no performance implications. In
followup CLs, we will start taking advantage of this flexibility.
Change: 146127565
|
|
|
|
|
|
|
|
|
| |
ParseSingleSequenceExample.
The attribute parsing code was moved over verbatim to example_proto_helper.*.
Change TF_CHECK_OK to TF_ASSERT_OK in all shape inference tests under core.
Change: 130041336
|
|
|
|
|
|
| |
- Prevent unnecessary tensorflow.Example copies when this method
is used in serving.
Change: 125788885
|
|
|
|
| |
Change: 123900938
|
|
|
|
|
|
|
|
|
|
| |
- Extract the fixed length and variable length feature configurations
output tensor names from a given GraphDef.
- This will allow for the use case of bypassing an unnecessary tensorflow.Example
serialize/deserialize at serving/inference time by extracting the configuration,
running the proto -> tensor helpers directly and feeding the graph with
the properly named tensors
Change: 122636456
|
|
|
|
|
|
| |
- Pre-allocate vector sizes when possible
- Maps added unnecessary overhead
Change: 120862895
|
|
- Methods added in core/util/example_proto_helper.{h,cc}
- This is in preparation for supporting a serving use-case
in which we can bypass an unnecessary serialize/deserialize
of Example protos by converting to dense/sparse Tensors directly.
- Note that this change is also tested by the python parsing_ops_tes
Change: 120608058
|