Changes from upstream/master #2
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
category: build
category: C API
category: CI
github_actions
category: Core
category: CPP API
category: CPU
category: dependency_changes
category: docs
category: GPU
category: NPU
category: IE Tests
category: inference
category: packaging
category: Python API
category: JS API
category: TEMPLATE
category: TF FE
category: TFL FE
category: PyTorch FE
no-match-files
labels
### Details: - Fix for #27903 - Update NF4 precision conversions in NPU plugin - *...*
### Details: - *static strided_sliced crop to use tensor* ### Tickets: - *160254*
### Details: - Added common base class for plugin configuration and migration in GPU plugin. - New config defines options as typed members of config class which shall provide more efficient access to the value than AnyMap - Property setting is possible before finalization only which shall localize the point when config become immutable. - All config options may be read from environment as well as from json config file when OV is built w/ debug caps. - Debug and release options are defined and handled similarly now --------- Signed-off-by: Vladimir Paramuzov <[email protected]>
### Details: - ISTFT-16 specification ### Tickets: - 159378 --------- Co-authored-by: Mateusz Mikolajczyk <[email protected]>
### Details: - *Implement reference conversion from fp16 to f8e4m3, and apply 2-step conversion, i.e., apply fp32->fp16, then fp16->fp8 for conversions from fp32 to fp8.* ### Tickets: - *[CVS-160375](https://jira.devtools.intel.com/browse/CVS-160375)*
Details: Microsoft Contrib Operator "FastGelu" for ONNX RT Tickets: N/A --------- Co-authored-by: Georgy Krivoruchko <[email protected]>
This PR is doing some optimization work on onnxfrontend com.microsoft.MatMulNbits operators with this changes: 1. it disabled const folding with use 75GB for phi3 INT4 model and 200+GB for llama3 INT4 model. 2. it trigger oneDNN matmul primitives, much benefits the GPU performance we tested this changes along with another PR #28163 , and confirmed phi3/llama3 INT4 model run well in LNL. --------- Co-authored-by: Yu, Zijun <[email protected]>
Details: Microsoft Contrib Operator "QLinearAveragePool" for ONNX RT Tickets: N/A --------- Co-authored-by: Georgy Krivoruchko <[email protected]>
### Details: - The original PR (#28788) has been mistakenly force-merged due to a mistake in merge queue settings. It was later reverted, so this is the "new" Ref PR. - Add reference implementation - Add tests ### Related PRs: - #28103 - #28698 - #28979 - #28999 ### Tickets: - CVS-158917 --------- Signed-off-by: p-wysocki <[email protected]> Co-authored-by: Roman Kazantsev <[email protected]> Co-authored-by: Pawel Raasz <[email protected]> Co-authored-by: Katarzyna Mitrus <[email protected]>
### Details: - Update onednn for gpu with latest 3.7 branch which includes ARL-H perf improvement ### Tickets: - *ticket-id*
### Details: - *Some customers can set policy ProcessDynamicCodePolicy to prohibit dynamic code. So need to disable JIT in such cases.* ### Tickets: - *162440*
### Details: - we have some nodes which is target to be optimized during runtime, the memory status can be corrupted among iterations for these nodes ### Tickets: - CVS-161830 --------- Signed-off-by: fishbell <[email protected]>
…#29058) Make an additional conversion from std::string to std::wstring at uv::util::to_file_path function when OPENVINO_ENABLE_UNICODE_PATH_SUPPORT was set Tickets: [162461](https://jira.devtools.intel.com/browse/CVS-162461) [146659](https://jira.devtools.intel.com/browse/CVS-146659)
Signed-off-by: Alina Kladieva <[email protected]>
Required for internal tests using requirement files packed here Signed-off-by: Alina Kladieva <[email protected]>
### Description: As CPP17 is the default standard of OpenVINO, there is no need to use the self-implemented optional struct. Note: discussed in #28942. ### Related PRs: - #28942 Co-authored-by: Michal Lukaszewski <[email protected]>
+ Add reorder to change input format from blocked to plain for dGPU to select onednn kernel for gemm. ### Tickets: - *162459*
Implicit casting error when different input and output data type. Use in_type instead of out_type during FUSED_OPS_VEC. ### Tickets: - *160534* --------- Signed-off-by: hyunback <[email protected]>
### Details: - *Use the new ZE_MUTABLE_COMMAND_EXP_FLAG_GRAPH_ARGUMENTS flag -> if the flag is still not supported by the driver, fallback to the deprecated one* - *Create fences or events only if they are used* - *Check Major version correctly* - *Make sure the Command queue is kept alive until the Fence is destroyed* - *Removing unused parameters from compiler adapter* --------- Signed-off-by: Bogdan Pereanu <[email protected]>
By utilizing the fact, that the base edge is always the first edge in the cluster. This helps to avoid unnecessary looping when updating the referencing edges. Also allocated dynamic output edges before the regions are created, so there is no need to create those regions and then removed them anyway.
Reverts #29031 because it introduced sporadic failures (e.g.: https://github.com/openvinotoolkit/openvino/actions/runs/13432909164/job/37528579365?pr=29086) The licenses whitelist should not contain AND expressions (see actions/dependency-review-action#842)
### Details:
After switching from ModelOptimizer to OVC, the order of
applying PrePostProcessing and MOCTransformations has changed:
MO path : [fw model conversion -> PrePostProcessing -> MOC] -> nncf
OVC path: [fw model conversion -> MOC] -> PrePostProcessing -> nncf
Since nncf is applied with a not fully optimized model, extra FQ ops
might appear,
which can affect both accuracy and performance.
e.g. Mul -> Conv fusion is not applied due to extra FQ
<img width="165" alt="{C6E93F2C-2CE3-4596-8D7F-ED7BD8013603}"
src="https://github.com/user-attachments/assets/3cbe6e07-9c07-4002-8b4c-9fb5bc662421"
/>
PrePostProcessing is not part of OVC, so we have to insert additional
Transformation calls inside PrePostProcessing.
### Tickets:
- *CVS-160786*
- CVS-161724
---------
Co-authored-by: Andrei Kochin <[email protected]>
Co-authored-by: Andrii Staikov <[email protected]>
### Details: - Mark `ov::element::undefined` as deprecated - Use dynamic type instead of undefined ### Related PRs: - openvinotoolkit/openvino_contrib#945 - #28524 ### Tickets: - CVS-161637 --------- Signed-off-by: Pawel Raasz <[email protected]> Signed-off-by: Raasz, Pawel <[email protected]>
) ### Details: - Add utils to core to set names on model input/output tensors - Define default tensor names and port separator - Use auto naming in model created by python API. Model outputs will have got names also. - The auto naming can be part of Model API but requires to add validation tools to find and resolve names collision - Auto naming cannot be added to Model creation in C++ as it can have impact on (de)serialization, requires more investigation ### Tickets: - CVS-159997 --------- Signed-off-by: Raasz, Pawel <[email protected]>
### Details: - *item1* - *...* ### Tickets: - *ticket-id* Co-authored-by: Mikhail Ryzhov <[email protected]>
### Details: - *Reverted `reset_state()` change (no changes regarding `reset_state()` is presented here regardless of branch name)* - *Added initialization of tensors in LLMInferRequest* - *Fixed copy of attention_mask* ### Tickets: - *ticket-id*
### Details: - *Return default NonZero type where possible. This allows to merge two NonZero nodes.* ### Tickets: - *CVS-160578* Signed-off-by: Maxim Vafin <[email protected]>
### Details: - *Extend mean_iou metric in single-image-test for semantic segmentation networks that have ArgMax at the end.* ### Tickets: - *EISW-158075*
### Details: - *Fix turbo mode* ### Tickets: - *E#156694* Signed-off-by: Bogdan Pereanu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.