| Commit message (Collapse) | Author | Age |
... | |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Waits until we are holding the GIL to lazily dereference the objects.
Session benchmarks, before:
benchmark_session_feed_direct_4B 1 0.000151
benchmark_session_feed_direct_4MB 1048576 0.000633
benchmark_session_fetch_direct_4B 1 0.000107
benchmark_session_fetch_direct_4MB 1048576 0.000481
benchmark_session_feed_grpc_4B 1 0.000307
benchmark_session_feed_grpc_4MB 1048576 0.000973
benchmark_session_fetch_grpc_4B 1 0.000192
benchmark_session_fetch_grpc_4MB 1048576 0.000876
After:
benchmark_session_feed_direct_4B 1 0.000150
benchmark_session_feed_direct_4MB 1048576 0.000153
benchmark_session_fetch_direct_4B 1 0.000106
benchmark_session_fetch_direct_4MB 1048576 0.000507
benchmark_session_feed_grpc_4B 1 0.000305
benchmark_session_feed_grpc_4MB 1048576 0.000360
benchmark_session_fetch_grpc_4B 1 0.000225
benchmark_session_fetch_grpc_4MB 1048576 0.000873
So feed_direct_4m is ~3x faster and feed_grpc_4m is ~2.5x faster
Change: 151170284
|
|
|
|
|
|
| |
least partially address #7464.
Change: 151168988
|
|
|
|
| |
Change: 151166174
|
|
|
|
|
| |
transposes and reshapes that feed reductions.
Change: 151162327
|
|
|
|
|
|
|
| |
- The resource path should not use the platform's file separator.
- Restore the commands to download and extract the Windows native libraries.
- Unrelated tweaks the README
Change: 151161421
|
|
|
|
| |
Change: 151158189
|
|
|
|
| |
Change: 151154030
|
|
|
|
|
| |
Use specialized GPU kernels on tensors when the permutation can be reduced to {0,2,1}, {2,1,0} or {1,0}.
Change: 151147354
|
|
|
|
| |
Change: 151147266
|
|
|
|
| |
Change: 151130707
|
|
|
|
| |
Change: 151128714
|
|
|
|
| |
Change: 151103646
|
|
|
|
|
|
|
|
|
|
|
| |
The general idea is that the scripts here download release artifacts (Java
sources, JNI native libraries) built using bazel as part of the TensorFlow
release process and packages them into a maven repository.
See README.md for instructions and layout of the artifacts.
Final step for #6926
Change: 151096674
|
|
|
|
| |
Change: 151087060
|
|
|
|
| |
Change: 151086446
|
|
|
|
|
| |
inference or training
Change: 151086011
|
|
|
|
| |
Change: 151077399
|
|
|
|
| |
Change: 151077249
|
|
|
|
|
| |
from other consts.
Change: 151075063
|
|
|
|
| |
Change: 151074297
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The previous range selection heuristic tended to be too aggressive in expanding
the range of the y-axis when dealing with values close to 0. For example, if
all of the values ranged between 0.65 and 0.7 then it would set the y-range to
somewhere near [-0.1, 0.7], making it impossible to see changes over time. If
all the values [0, 1e-4] then it would select a range close to [-0.1, 1e-4].
This change adds logic to avoid expanding the range of the y-axis too much. It
also modifies the 'ignore outliers' logic to select a range that includes 90%
of the data, as opposed to the checked-in version that uses 80%.
Change: 151071239
|
|
|
|
| |
Change: 151071183
|
|
|
|
|
|
|
| |
now requires named arguments in order to prevent mixing up labels and logits
Also added lint fixes and fixed a broken/hidden test that wasn't being run
Change: 151069880
|
|
|
|
| |
Change: 151065650
|
|
|
|
|
|
| |
- Renaming SDCARegressor to SDCALinearRegressor
- Improving documentations
Change: 151065563
|
|
|
|
| |
Change: 151064926
|
|
|
|
| |
Change: 151064589
|
|
|
|
|
|
| |
in test cases, and switch the size of pooling_ops_3d_test back to "small". Delete the Extract3dPatch function which is not used anywhere anymore.
Change: 151062901
|
|
|
|
|
|
|
| |
with exceptions, use longs for dimensions, and simplify method signatures where possible.
Note this is a breaking change and will cause the TF demo in tensorflow/examples/android to not work with older jars/.so binaries built from tensorflow/contrib/android
Change: 151061702
|
|
|
|
|
| |
Add "Any" to client library routines and add tests for it.
Change: 151060097
|
|
|
|
| |
Change: 151050903
|
|
|
|
| |
Change: 151050198
|
|
|
|
| |
Change: 151049836
|
|
|
|
|
|
| |
threshold. Reenable conv_ops_3d_test.
Change: 151049454
|
|
|
|
| |
Change: 151049375
|
|
|
|
| |
Change: 151048951
|
|
|
|
| |
Change: 151048554
|
|
|
|
| |
Change: 151047342
|
|
|
|
| |
Change: 151046259
|
|
|
|
| |
Change: 151042735
|
|
|
|
|
| |
These were producing a warning popup when you run the notebook in the current jupyter (4.3.0).
Change: 151041728
|
|
|
|
|
|
| |
parse_single_example.
Change: 151038761
|
|
|
|
| |
Change: 151036764
|
|
|
|
| |
Change: 151036671
|
|
|
|
| |
Change: 151031162
|
|
|
|
| |
Change: 151031147
|
|
|
|
| |
Change: 151018201
|
|
|
|
| |
Change: 151017665
|
|
|
|
| |
Change: 151015024
|
|
|
|
| |
Change: 151012934
|