Port from https://github.com/openvinotoolkit/openvino/pull/22881
* Created new article about Release Policy,
* Moved System Requirements Article under Release Notes.
* Removed environment variables info from torch.compile article.
Jira: 132590
### Details:
- *Add NPU in the supported hardware matrix*
- *Add NPU in the list of supported plugins*
- *Add NPU plugin in CONTRIBUTING.md*
### Tickets:
- *EISW-119403*
PR on master: https://github.com/openvinotoolkit/openvino/pull/23919
**Details:** Recommend openvino-tokenizers for some operations in case
failed conversion
**Ticket:** TBD
Port of https://github.com/openvinotoolkit/openvino/pull/23878 from
master
Signed-off-by: Kazantsev, Roman <roman.kazantsev@intel.com>
### Details:
- *Added `kornia` dependency for lightglue model*
- *Disable GFPGAN since it doesn't support latest `torchvision`*
- *Remove `transfo-xl-wt103` as not supported on latest `transformers`*
- *Remove XFAIL from `facebook/levit-128S`*
### Tickets:
- *ticket-id*
### Details:
- Used actual shapes for Tensors creation.
### Tickets:
- CVS-137149
---------
Co-authored-by: Michal Lukaszewski <michal.lukaszewski@intel.com>
### Details:
- Shapes inference can result into an appearance of the zero shape
memory on the node inputs. Such nodes are considered as non-executable,
so no prepareParams are called for them, thus no
executors are created. So, it is necessary to avoid setting numa id for
non-executable nodes.
- Currently seg fault happens when node gets zero shape memory on the
inputs.
Issue introduced in scope of:
- https://github.com/openvinotoolkit/openvino/pull/23593
### Tickets:
- *CVS-137763*
### Details:
We have introduced a new custom action that uploads and downloads cached
files as monolithic tarballs. It functions similarly to the GHA cache
action but utilizes mounted shared disks to save and restore caches.
The action searches for tarball files in the specified cache directory,
filters them by restore keys, and sorts them by access time. The most
recently accessed file will be downloaded and extracted to the runner
file system. As a post-action step, it archives files to the tarball and
uploads them to the remote share. Additionally, it provides the ability
to clean up old cache files.
It also includes the following separate actions:
- Restore
- Save
- Cleanup
These actions can be used individually in workflows
For example, its usage is demonstrated in the PR, which includes changes
to the .github/workflows/linux_riscv.yml file
### Tickets:
- *ticket-id*
### Details:
- Compressed int4 weight is now supported with OneDNN
- transpose.cpp change for int4 support from ngraph transform is a
temporal solution. To be handled from GPU plugin.
### Tickets:
- 124115
PagedAttention operation exposed in Python API only for easier vLLM
openvino integration. It is not intended to be used outside our
integration work in vLLM and similar applications where we can use
PagedAttention. Exposed as a hidden part of API, will not be documented.
Connected to already existing implementation in CPU plugin.
Operation is not a part of any public opset.
### Details:
- `ov::test::utils::compare(...)` used for output vs reference values
comparison (where possible).
- Tests with a need of complex changes were marked with `legacy_compare`
flag. These tests still use legacy raw data comparison and theirs
transition will be addressed in further PR's.
### Tickets:
- CVS-135274, CVS-135275
---------
Co-authored-by: Michal Lukaszewski <michal.lukaszewski@intel.com>
### Details:
- Edited and overloaded the __getitem__ operation for `Shape` and
`PartialShape`
- Added tests for correct the implementation
### Tickets:
- 75641
- #21968
---------
Co-authored-by: Jan Iwaszkiewicz <jan.iwaszkiewicz@intel.com>
Co-authored-by: Przemyslaw Wysocki <przemyslaw.wysocki@intel.com>
### Details:
Fix segfault in Constant when destroy not initialized string buffer:
- Initialize buffer in Constant before exception throw to avoid
segfault.
- Improve default ctor of` StringAlignedBuffer`.
- Set only required members in `SharedStringAlignedBuffer`'s dtor.
- Add unit test for `StringAlignedBuffer`.
- Use template `get_ptr` instead cast.
- Optimize assertion error message.
### Tickets:
- Relate to
[CVS-119224](https://jira.devtools.intel.com/browse/CVS-119224)
### Details:
- *Extend Base test class by default threshold calculation based on
whole model*
- *Choose max values as thresholds by default*
- *Create map to add custom threshold per operation*
### Tickets:
- *[CVS-136744](https://jira.devtools.intel.com/browse/CVS-136744)*
Jan Iwaszkiewicz