* have Im2ColNd support all types and allow customized padding value.
* only specialize the template in order NCHW.
* fix build break.
* fix build break
* Implement N-gram
Do not load unnecessary pool n-grams. Add String typed tests.
Set output size to the mav ngram_index value plus 1.
* Address security warnings and some review comments.
* Fix build issues, rework sampling to try all n-gram sizes at a given offset.
* Rework the loop so all n should be tried at a given offset
and we do not add the same items all over again such as
b,c and next we try b,c,d but we no longer add b,c again.
* Compute hash incrementally so we do not re-hash elements that were
already there when we add more elements to n-gram.
* Address review comments.
TODO: Remove all attribute.
* Remove all attribute, adjust tests. Correct docs.
* Address more review comments.
* Create Type And Shape inference function.
* Address review comments. Implement batch mode per new spec.
* Correct switch bracing in OutputResult and re-test.
* Fix shape error message within TypeAndShapeInferenceFunction.
Implement Inverse for hyberbolic ops
Eigen will add support for asinh, acosh and atanh in the upcoming release. But until then for completeness of opset9 we have std based implementation.
* merge function compile interface
* fix build error
* fix linux build break
* fix static cast issue; fix clang style
* fix argument change
* use alignment allocation;fix comments in pr
* fix linux break
* apply clang format
* rename according to comments in pr
* rename according to pr comments;remove useless file
* remove the need_compile flag
* avoid passing whole session state
* Implementation of OneHot op
* Implementation of OneHot op
* Implementation of OneHot op.
* one hot op
* onehot
* OneHot op.
* Disable some Eigen related warnings
* fix build
* Add comment for potential optimization opportunity using Eigen threads
* Remove this op from the broken tests
* More API changes, remove 'Inference' from function names. Remove enum values. Make Status match other types.
* Switch to bool instead of int, and remove stdbool
* added end-to-end nuget package test
* reset the changes in OnnxRuntime.CSharp.proj
* revert the testdata directory path
* revert inference tests proj file
* added script for running end-to-end tests
* fix in the runtest.bat
* added error checking in runtest
* fixed paths in the test project
* added runtest.sh
* fix protoc path
* updated executable attributes for the runtest scripts
* added some log to debug protoc failures
* removed the protoc and duplicate test code, reuse unit-test code for end-to-end test
* copy always
* fix working dir paths in runtest.sh
* added a build.py flag to download test data without running the c++ tests
* added a script for running the test under docker
* added script for docker run of the test
* mkldnn sum derives cpu sum_6. removed final
* MKLDNN Relu, Sum and BatchNorm
* PR review changes
* more review changes
* relu primitive typo corrected
* run bw and fw sequentially for GRU if using MKLDNN
* word conv embedding custom op
* run bw and fw sequentially for GRU if using MKLDNN
* Add word conv embedding custom op
* fix build break in linux
* fix macos build break
* resolve the comments
* refine the comments
* remove unnessary comment
* rename the function to calculate the length of eache word in a sequence
* add license info and fix typo
* Minor updates to exception message
* update models folder to new location
* update copy to preservenewest
* reenable pretrained test
* added some debugging info for build
* update pretrained test, and tensor proto definition
* support hyperbolic fns
This commit adds support for sinh and cosh. Support for hyperbolic inverses is not available in Eigen yet.
* Make constructors explicit
* remove tests from exclude list
* Revert "remove tests from exclude list"
This reverts commit 2112a30b57d5a899991de4847e948e700a44e85d.
* remove test names from excluded list
* remove tanh since its already implemented
- apply any transforms to the main graph and any subgraphs first
- call Graph::Resolve() once on the main graph, which will recurse into the subgraphs
- previously it was called after the transform on each subgraph, which results in it traversing up to the main graph to call resolve, and that resolve call recursing into all subgraphs every time.
This avoids lots of unnecessary Graph::Resolve calls, and prevents subgraphs from being broken by SessionStateInitializer::InitializeAndSave calling graph_.CleanAllInitializedTensors() prior to final Graph::Resolve call. If a subgraph has optional inputs the backing initializers were removed by CleanAllInitializedTensors causing the next Resolve to incorrectly turn them into required inputs.
