* Implement separator tokenizer with TST.
TODO: Clarify what to do if the output is empty and no start/end text
markers required. Also see if the current search algo is acceptable.
* Add utf8 util test
* For empty output produce [C] -> [C][0], [N][C] -> [N][C][0]
* Augument TST search with match conflict resolution in favor of
earlier specified pattern matches.
* Address MAcOS build error.
* Adjust error message
* Address review comments.
* Remove nested loops.
* Remove 3rd party utf8 validation code.
* Address review comments part I.
* Move padding outside start/end markers.
Split unit tests for invidividual test cases.
* Fix a common prefix bug reported by Xavier.
* Upgrade gpu build to CUDA 10 + cudnn 7.3
* update the yaml file for python package building
* switch to the cuda9.1 docker file if the CUDA_VER is cuda9.1-cudnn7.1
* Add build step to remove the cuda msbuildcutomization file after build, otherwise, the cuda high version could impact the lower version build
* update vs path
* update the path
* Update Dockerfile for ARM build and build description
* Update BUILD.md for ARM build
* Fix typo
* Kezhan/renaming graph_base.h to graph.h (#95)
* rename graph.h to graph_viewer.h
* rename graph_base.h to graph.h
* Implement StringNormalizer (#69)
* Imlpement StringNormalizer
Add mixed language tests, test case insentive path.
* Create a locale on the fly. Default locale does not seem to create well.
* Add CI language-pack-en to make default locale available.
Catch and translate locale creation exception to make the message
meaningful.
* Make sure locales are configured on Ubuntu.
* Split the CI pipelines (#94)
* Add the split build badges (#96)
* Split the CI pipelines
* Update build bdge display table in README
* Revert Softmax optimizations using openmp. (#97)
Revert "Simpler unused parameter in #if defined() switch."
This reverts commit 6b00e6bb4d.
Revert "Better opemmp parallel group count calculation in Softmax parallel running."
This reverts commit c530064ebe.
Revert "Optimize softmax cpu by parallel using openmp."
This reverts commit e7bdfa00db.
* Delete logo as per marketing requirements (#98)
* Update Dockerfile for ARM build and build description
* Update BUILD.md for ARM build
* Fix typo
* Eliminate the confusing double negative
I was having trouble parsing the caveat NOTE, proposing wording changes that I think reflect the meaning and avoid the confusion.
* Eliminate double negative without further explanation on role of this file.
Incorporate @pranavsharma feedback.
Revert "Simpler unused parameter in #if defined() switch."
This reverts commit 6b00e6bb4d.
Revert "Better opemmp parallel group count calculation in Softmax parallel running."
This reverts commit c530064ebe.
Revert "Optimize softmax cpu by parallel using openmp."
This reverts commit e7bdfa00db.
* Imlpement StringNormalizer
Add mixed language tests, test case insentive path.
* Create a locale on the fly. Default locale does not seem to create well.
* Add CI language-pack-en to make default locale available.
Catch and translate locale creation exception to make the message
meaningful.
* Make sure locales are configured on Ubuntu.
Fix the bug for RNN/GRU/LSTM auto fall back, the supported activation should cover bidirectional mode.
Update the test to remove the flag to enable gpu test since we can auto fallback to cpu
* Support fusing 3D Conv with Add/Mul.
With this PR, the subgraph 3D Conve->Add->Mul in Resnet3D can be fused into one 3D Conv.
* Updated it based on feedback.
* Updated it based on review feedback.
* Change the implementation of scale_by_axis rather than Mul.
* Refactor the code to make the compiler optimize it easily.
