IntenseWebs/openvino
3
0 Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
13,746 Commits 38 Branches 54 Tags
6246fb0c23cef91004fe309962dceaf484a6feb8
Commit Graph
13746 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Nikolay Shchegolev 6246fb0c23 [CPU] StringMemory for the output tensor in the InferRequest (#21746) 2023-12-19 15:20:15 +04:00
Roman Lyamin 6060683f4c [GPU] Fix for is_user_cpu() function (#21744) 2023-12-19 10:48:13 +01:00
Sofya Balandina 8f13219728 [op conformance] Made fixes to allign with accuracy validation (#21347) 2023-12-19 13:43:40 +04:00
Xiping Yan a3a3f0b55c [CPU] Fix chatglmv2 accuracy issue and crash with input 1024 token in SPR (#21360) 2023-12-19 13:13:35 +04:00
Pawel Raasz 63cb89fe1b Use move instead copy (#21732) 2023-12-19 12:43:28 +04:00
Maciej Smyk e94feea5c7 [DOCS] Update of graphs in repo (#21737) 
* csv files

* Update GNA.rst

* csv files

* csv files

* csv

* csv

* removal of duplicates

* duplicates removal

* Delete small_IR_graph_demonstration.png

* moving graph files to static folder

* Update LowPrecisionModelRepresentation.rst

* update
 2023-12-19 09:20:24 +01:00
Anastasiia Pnevskaia 3c157b2a1b [TF FE] [Telemetry] Added error_info event to TF FE. (#21736) 
* Added error_info event to TF FE.
 2023-12-19 12:14:36 +04:00
Pawel Raasz 499a9694e5 Fix Coverity CID 1529136, 1529124 (#21733) 
* CID 1529136, fix auto makes a copy of PartialShape
- minor updates to reduce redundant Rank object creation

* CID 1529124 fix copy caused by auto
 2023-12-19 07:06:33 +00:00
rghvsh a4e8f9d98c [ONNX] Extend ONNX Frontend with BitwiseAnd-18 operator (#21585) 2023-12-19 09:54:55 +04:00
Vladimir Paramuzov 63b23a1ebb [GPU] Enable state precision conversion to infer precision (#21729) 2023-12-19 09:36:03 +04:00
Min, Byungil 0e92469330 [GPU] bugfix slice mem calculation for batch iteration (#21716) 
Signed-off-by: Min, Byungil <byungil.min@intel.com>
 2023-12-19 14:20:38 +09:00
Taylor Yeonbok Lee b770780b89 [GPU] Skip redundant gather in stateful model (#21681) 
* Skip redundant gather in stateful model

* Fix memory reuse issue for node skipped at runtime.
If the node is not marked as can_be_optimized at build time, memory dep is not properly applied
=> So it can cause the wrong memory reuse
 2023-12-19 00:00:39 +00:00
Jan Iwaszkiewicz c59498be9c [PyOV] Allow single inputs in form of lists of simple types (#21734) 2023-12-18 22:14:31 +01:00
Sergey Shlyapnikov 0e496fa0b5 [GPU] Include Select node to shape_of sub-graph and add CPU impl to avoid runtime synchronizations (#21738) 2023-12-18 19:52:11 +00:00
Wanglei Shen d5b80062d9 support E3950 in MT 2.0 on both Linux and Windows (#21735) 2023-12-18 20:42:22 +01:00
Gorokhov Dmitriy 058d488083 [CPU] FullyConnected acceleration with 4bit weights decompression on SPR (#21723) 2023-12-18 23:16:45 +04:00
Maksim Kutakov 18a9c772b4 [CPU] Plugin optimizations for LLM beam search via states (#21642) 2023-12-18 23:16:32 +04:00
Eddy Kim 01967bdbff [GPU] Adding a switch to change the caching policy (#21443) 
* added a new property to enable full model caching

* added a new public property cache_mode

* updated according to clang-format guide

* changed default caching policy to optimize_speed
 2023-12-18 17:50:33 +00:00
Eddy Kim 63e08f9965 [GPU] adding support for weights compression to gather (#21711) 
* 1st version

* fixed style check error

* added unit tests

* removed unnecessary comments

* updated hash to include decompression type

* applied code reviews

* applied code reviews

* fixed unit tests
 2023-12-18 17:46:11 +00:00
Mateusz Tabaka d03dc4fa2a Add LSTMCellFusion transformation (#21594) 
* Add LSTMCellFusion transformation

Partially fixes: CVS-125605

* code style

* fix accuracy issue

* add headers
 2023-12-18 18:35:25 +04:00
Mateusz Mikolajczyk ec8765f950 Fix L to not broadcast it (#21724) 2023-12-18 17:31:56 +04:00
Sergey Shlyapnikov 032ac898e2 [GPU] Apply is_non_decompression_multiply() callback only for compressed models (#21719) 2023-12-18 12:20:20 +00:00
Mateusz MikolajczykandAnastasia Kuporosova 98e8caad79 [PyAPI][Opset13] ScaledDotProductAttention-13 PyAPI improvements + tests (#21422) 
* Improve ScaledDotProductAttention PyAPI

* Add tmp tests

* Improve PyAPI & add tests

* Fix test issue

* Fix pytest type

* Update src/bindings/python/tests/test_graph/test_scaled_dot_product_attention.py

Co-authored-by: Anastasia Kuporosova <anastasia.kuporosova@intel.com>

* Improve const creation

---------

Co-authored-by: Anastasia Kuporosova <anastasia.kuporosova@intel.com>
 2023-12-18 12:35:14 +01:00
Piotr Krzemiński 8ff25cf4a2 [SPEC] Update ScaledDotProductAttention specification for broadcasting of batch dimensions (#21714) 
* [SPEC] Update spec for broadcasting of batch dimensions, improve clarity

* [SPEC] Update code block style, comments in xml, attention_mask details
 2023-12-18 12:19:06 +01:00
Pawel Raasz a6f3c7d1b2 Us dynamic dim in pattern if only one valid bound (#21694) 2023-12-18 15:02:01 +04:00
Mateusz MikolajczykandRoman Kazantsev a1e296eaa8 [TF FE][Opset13] Enable Multinomial operator in TF frontend (#20646) 
* Enable Multinomial operator in TF frontend

* Implement requested changes

* Update tests/layer_tests/tensorflow_tests/test_tf_Multinomial.py

Co-authored-by: Roman Kazantsev <roman.kazantsev@intel.com>

* Align with CPU implementation

---------

Co-authored-by: Roman Kazantsev <roman.kazantsev@intel.com>
 2023-12-18 12:00:04 +01:00
Tomasz Jankowski a29843013d [Ref] Fix ScatterElementsUpdate lowest value bounding (#21669) 
* Use lowest for value bounding

* Add test for lowest on float
 2023-12-18 11:57:01 +01:00
Nikolay Shchegolev 554e7d6730 [CPU] Dynamically shaped model with string tensor output fails (#21713) 2023-12-18 14:44:36 +04:00
Mateusz Mikolajczyk be476519fb [PT FE] ScaledDotProductAttention fix issue with other float dtypes (#21638) 
* Fix scale input

* Improve torch scaled dot tests
 2023-12-18 14:32:54 +04:00
Ivan TikhonovandIlya Lavrenov e224e4f4f0 Introduce OriginalPrecision attribute (#21696) 
* Introduce OriginalPrecision attribute

* resolve review comments

---------

Co-authored-by: Ilya Lavrenov <ilya.lavrenov@intel.com>
 2023-12-18 14:10:46 +04:00
Georgy Krivoruchko 62c9469f52 [ONNX] Reflected Shell-scripts to run on Windows (#18672) 
* Reflected Shell-scripts to run on Windows

* Fixes

* Removed script from tests_compatibility
 2023-12-18 13:51:20 +04:00
Eddy Kim 35dafce513 [GPU] move convert after gather to reduce memory usage (#21616) 2023-12-18 13:34:46 +04:00
Mateusz MikolajczykandMichal Lukaszewski 610e0fab5c [Opset13] ScaledDotProductAttention-13 input validation improvements (#21281) 
* Improve test naming to match op

* Initial draft for scaleddot shape infer

* Fix formatting

* Improve shape & type infer

* Add shape tests

* Fix issues in shape infer

* Fix type infer

* Improve type_prop tests

* Ignore attention when causal

* Fix template type shape infer

* Fix issues with scalar inputs + test improvement

* Fix `get_node_input_partial_shapes`

* Improve shape/type validation and tests

* Allow for broadcastable inputs

* Improve tests

* Add CPU shape infer test

* Add broadcast shape infer test

* Use const

* Use const &

* Improve query input handling

* Fix test issues

* Change broadcast rules

---------

Co-authored-by: Michal Lukaszewski <michal.lukaszewski@intel.com>
 2023-12-18 10:30:54 +01:00
Wang Wangwang 3a2958b360 [GPU] Fix convert fc to compressed when the input is SubGraphOp (#21687) 2023-12-18 13:29:16 +04:00
Vitaliy UrusovskijandIlya Lavrenov e71707b7d6 Remove extra includes (#21699) 
Co-authored-by: Ilya Lavrenov <ilya.lavrenov@intel.com>
 2023-12-18 13:20:54 +04:00
Irina Efode d613bb0e65 [CPU][ARM] Fix build CPU ARM plugin (#21722) 2023-12-18 13:20:33 +04:00
Piotr Krzemiński acb3f5ef28 [Core][Tests] Add attribute tests and decompotion tests for Scaled Dot Product Attention (#21356) 
* [CORE][TESTS] Add attribute tests and decompotion tests for scaled dot product attention

* Update batch_norm_decomposition_test.cpp

* Clang fix

* Update scaled_dot_product_decomposition_test.cpp

* [TESTS]STS] Fix CIs with rename and clang fix

* [TESTS] Fix decomposition for CIs

* [TESTS] Remove constant mode

* [TESTS] Fix visitor test for CI

* [Tests] Add suggested broadcast tests
 2023-12-18 10:18:48 +01:00
Jan Iwaszkiewicz 57b1916f8d [PyOV] Remove restrictions on opset13.constant to align with opset1-12 (#21674) 2023-12-18 10:02:43 +01:00
Sebastian Golebiewski a9a935ee6e [DOCS] Fix notebooks sections in menu (#21695) 
* Fix notebooks sections in menu

* prepend commands in cmake
 2023-12-18 13:01:22 +04:00
Mateusz MikolajczykandRoman Kazantsev 1783aad48d [PT FE][Opset13] Add translation for torchscript aten::multinomial operator (#20397) 
* Enable aten::multinomial in PT FE

* Add aten::multinomial test

* Fix typo

* Update tests/layer_tests/pytorch_tests/test_multinomial.py

Co-authored-by: Roman Kazantsev <roman.kazantsev@intel.com>

* Update multinomial to changes in specification

* rm ;

* Add temporary convert for multinomial i64 support

* Revert convert

* Fix torch test

---------

Co-authored-by: Roman Kazantsev <roman.kazantsev@intel.com>
 2023-12-18 09:54:06 +01:00
Georgy Krivoruchko 0972de4da8 [ONNX] Refactoring tests on API 2.0 (#21712) 
* Moved test in onnx_import_com_microsoft to new API
* Updated onnx_import_constant_folding.in.cpp to new API
* Tests in onnx_import_controlflow.in.cpp moved to new API
* Tests in onnx_import_convpool.in.cpp moved to new API
* Test in onnx_import_deprecated.in.cpp moved to API 2.0
 2023-12-18 11:31:32 +04:00
Ivan Tikhonov c4a49a3987 Add reset_state method to Python API (#21660) 
* Add reset_state method to Python API

* fix missprint

* codestyle

* extend python tests

* delete debug code

* resolve review comments
 2023-12-18 11:26:53 +04:00
Roman Lyamin 0e7248430c [Transformations] Support precision conversion for PriorBox (#21702) 2023-12-18 11:19:59 +04:00
Jade Cho 81f336b352 [GPU] Fix the mismatch between format and shape. (#21381) 
+ Change broadcast type of broadcast node which is added for binary add fusion.
 2023-12-18 15:51:47 +09:00
Chenhu Wang 6aa0a6677d [SnippetS] Binary size reduce, executor cache disabled if enable perf count (#21494) 2023-12-18 09:53:43 +04:00
Sungeun Kim da662116d4 parameter's event will be included execution args (#21377) 
- all primitives generate event.
- remove reorder from test-case
 2023-12-18 10:40:20 +09:00
Georgy KrivoruchkoandIlya Lavrenov 7b00e804de [ONNX] Refactoring tests on API 2.0 (#21707) 
* Removed usage of API1.0 in onnx_reader_external_data.cpp

* Removed API1.0 usage in onnx_import.in.cpp

* Replaced old calls by convert_model in onnx_import.in.cpp

* Removed a deprecated API in onnx_import.in.cpp

* Fixed code style

---------

Co-authored-by: Ilya Lavrenov <ilya.lavrenov@intel.com>
 2023-12-17 13:44:38 +04:00
Georgy Krivoruchko 58a3debaa7 Try to fix download issue (#21710) 2023-12-17 12:31:01 +04:00
Eddy Kim e109f735dc [GPU] faster model caching (#21479) 
* faster large data loading by double buffering

* added get_allocating_order() that is cached for faster loading

* added offset usm memory copy tests

* removed negative_same_names_within_different_networks_cache

* replaced get_processing_order() with processing_order
 2023-12-16 04:50:58 -08:00
Ivan TikhonovandMaksim Kutakov fc88e14e83 Update evaluate methods of ReadValue/Assign operations (#21372) 
* Update evaluate methods of ReadValue/Assign operations

* Update evaluate methods of ReadValue/Assign operations

* Prepare IVariable interface

* Support initializing subgraph

* Add exception types to the edge conflict searching algo

* Add test for init graph edge conflict

* Remove ShapeOf from consumer nodes in memory conflicts detection algo

* update template plugin

* Fix template tests; synchronize reset flag between VariableState and VariableValue objects

* fix legacy (Assign/ReadValue v3 + CNN Network) tests

* fix python tests

* fix python tests

* resolve review comments, apply cpu patch

---------

Co-authored-by: Maksim Kutakov <maksim.kutakov@intel.com>
 2023-12-16 11:26:39 +01:00