Decompression attribute (that is present in models with FP16 precision)
prevents the weights to be constantfolded. Weights constantfolding is
required by CompressQuantizeWeights to compress the weights to low
precision format.
Ticket: CVS-117310
* Shared ShapeOf transformation via SharedOpOptimization. Allow EliminateGatherUnsqueeze pattern to have intermediate binary operation. Changes covered with transformation tests
* Move Reshape up through binary op to be optimized by NopElimination transformations
* Optimizes Shared Convert
* SharedOpOptimization: give higher priority for nodes of later versions to become root nodes
* Deleted old transformations
* Added revalidation for modified nodes
* Added binary op revalidation to the EliminateGatherUnsqueeze in order to have correct shapes for other matchers in the same graph rewrite
* Made PrepareShapeOpsForEliminationAroundBE independable of the upper node
* Introduces GroupedSliceToVSplitOptimization optimization
* Preserve output names during GroupedSliceToVSplitOptimization
* Revert "Made PrepareShapeOpsForEliminationAroundBE independable of the upper node"
This reverts commit 96785b24c9.
* Comments are adressed
* SharedOpOptimization: removes Base classes from the rules
* ResolveNamesCollisions transformation refactoring; enable it in MOC
* fix the description
* call ResolveNamesCollisions transformation in the frontends; resolve review comments
* Resolve review comments
* fix EliminateUnsqueezeGather and AlignMixedTypes transformations
* Add subgraph body comparison
* Avoid confusing function name
* Skip failing snippet test
* Skip some ov_snippets_func_tests
* Derive comparison flags
* Skip snippet test
* Drop on bodies mismatch
---------
Co-authored-by: Michal Lukaszewski <michal.lukaszewski@intel.com>
Co-authored-by: Ivan Tikhonov <ivan.tikhonov@intel.com>
* Add static shape adapter
- Adapters holds CPU dimension which can be reference to it or vector
- Add ov::optional for holding optional result from shape inference
- Add new `infer` function in `IStaticShapeInfer`
* Temporary support of StaticShape
* Minor corrections in ShapeInferenceTA
* Migrate shape_infer to new interface version
* Replace StaticShape by adapter implementation
* Replace IShapeInferCommon by IStaticShapeInfer
* Correct code formatting
* Fix build issues
* NodeValidationFailure::create for StaticShapeRef
* Handle sequence ops with non constant W, R, B inputs
Ticket: CVS-49207
* fix tests
* tests
* fix gna tests
* add include
* change type for seqLenIdx
* rename is_constfoldable to is_on_constant_path
---------
Co-authored-by: Ivan Tikhonov <ivan.tikhonov@intel.com>
* Provided visualization of partial values and labels. Adopted DimensionTracker for better equivalence tracking
* Addressed comments and fixed one test
* Use GroupNormalization in ONNX FE
* Add and enable decomposition transformation
* Add NodeRegistry
* Adjust eps type
* Use Unsqueeze for bias and scale adjustment
* Add transformation tests
* Add Squeeze for bias and scale to ONNX FE
* Simplify get_range helper
* Add cast double to float eps helper
* Adjust reduction axes
* Style alignment
* Add float eps values in tests
* Enable accuracy comparator in transformation tests
* Skip F16 ref test
* Namespace adjustment
* Const ref for shape vec
* Remove nested scopes from transform tests
* Rename function to model in tests
* Add more const to variables
* Use common input shapes and elem type in transform tests
* Add ticket number to the test
* Update tests headers and names
* Use compatible ror rank check
* Use move to insert splits in new shape
* Add docs to cast_eps helper
* Additional test for eps
* Use versions instead of opset
* Include each op separately
* rrn seq base common shape infer init
* Update and add shape infer to all seq rnn based ops
* Use shape_infer in lstm seq core op
* Use shape_infer in rnn seq core op
* Unified shape infer for rnn cell base
* Update and add missing cells shape_infer functions
* Use shpe_infer in rnn_cell
* Update shape_infer map
* shape_infer tests
* Move new shape_infer to TA map
* More tests and shape improvements
* Introduce common base tests
* Fix merge conflict
* Missing setters and default ctor tests
* Use RNNCellBase class instead of template type op
* Update v0 LSTMSequence to use RNNCellBase as parent class
* Style
* V0::LSTMSequence update
* Specify opset in shape infer registry
* Move rank check
* Output vec generation refactor
* Update num_dir_validation
* Tests warining fix
* Test types adjustment
* Commented code cleanup
* Move test helpers to test fixture
* Common default ctor tests for Cell
* Update GRU shape infer tests
* Update LSTM shape infer tests
* Update RNN shape infer tests
* File names update
* Functions renaming
* Cast hidden size in test
* Move v0::LSTMSequence conformance test
---------
Co-authored-by: Pawel Raasz <pawel.raasz@intel.com>
* fix transformations
* clang fix
* fix unit tests - check the both Pad versions
* add unit tests checking negative padding
---------
Co-authored-by: Ivan Tikhonov <ivan.tikhonov@intel.com>
* ResolveNamesCollisions transformation refactoring; enable it in MOC
* fix the description
* call ResolveNamesCollisions transformation in the frontends; resolve review comments
* Resolve review comments
* Rename ngraph with ov in common_test_utils
* Fix comments
* Delete opset includes from commonTestUtils
* Fix CI and code style
* Apply comments
* Fix CI and apply comments
* Added TSForwardBase class and a new transformation to reset no_sinking attribute
* Refactoring
* fix an issue with legacy_output_names
* resolve review comments
* Resolve review comments
* keep Const+DecompressionConvert pattern for CPU
* temporary disabled failing unit-tests
* disable CF by modifying bounds evaluate as well; minor corrections
* added TODOs with ticket numbers
* join const+decompression markings
* minimized convert_precision.cpp changes
* minor corrections
* refactor fp16 transformations: moved into separate fp16_compression folder
* style-fix
* minor fixes
* do not disable evaluate and CF in shape path
* safer disabling of Const conversion
* style-fix and minor corrections
* restore original placement of ConvertPrecision
Shuangji YangandIvan Tikhonov
Mateusz Tabaka
Evgenya Stepyreva
Anton Voronov
Pavel Esir
Pawel Raasz
Wang, Yang
Vladislav Golubev
Oleg Pipikin
Evgeny Kotov
Katarzyna Mitrus
Ilya Lavrenov
Ilya Churaev
Tomasz Dołbniak
Mateusz Bencer
Georgiy Manuilov