Sync with Microsoft ONNX Runtime - 15/09/2025 - #811
Merged
Conversation
…t#25571) ### Description Ensure all pre-allocated tensors do not trigger unnecessary data copying. e.g. the WebNN EP always binds its tensor to 'ml-tensor'. In such cases, the tensor ID might change after binding, but copying data for these tensors should still be avoided. ### Motivation and Context This improves efficiency and avoids redundant operations.
### Description * Relax tolerances for ONNX attention node tests ### Motivation and Context * Tests failing in Qualcomm Internal CI before relaxing tolerances, tests passing after
CPU MoE Kernel ``` name: SwigluMoEBlock, quant_bits: 0, dtype: FP32, batch: 1, seq_len: 16, max_diff: 2.682209014892578e-07 .name: SwigluMoEBlock, quant_bits: 0, dtype: FP32, batch: 1, seq_len: 32, max_diff: 2.980232238769531e-07 .name: SwigluMoEBlock, quant_bits: 0, dtype: FP32, batch: 2, seq_len: 16, max_diff: 2.980232238769531e-07 .name: SwigluMoEBlock, quant_bits: 0, dtype: FP32, batch: 2, seq_len: 32, max_diff: 4.172325134277344e-07 .MoE CPU kernel time: 15.721677541732786 ms . ---------------------------------------------------------------------- Ran 5 tests in 30.217s ```
### Description In latest WebNN spec, it requires the `scale` and `zeroPoint` rank to be same as the input rank for `quantizeLinear` and `dequantizeLinear` operators. ### Motivation and Context This PR adjusts the shape of scale and zeroPoint by appending or prepending 1s to their dimensions if they are not equal to the input rank to align with the latest spec. BTW, this PR also fixes some nits and makes some code improvement.
### Description Update Qnn default version to 2.38.0.250901
…time. (microsoft#26011) ### Description <!-- Describe your changes. --> Add ONNX patch with changes from onnx/onnx#7253 to avoid regenerating proto files and improve rebuild time. ### Motivation and Context <!-- - Why is this change required? What problem does it solve? - If it fixes an open issue, please link to the issue here. --> Decrease incremental build times during development.
- Making sure that the Uniquename function is not called repeatedly - Added unit test for int64 GatherND test for coverage ### Description GatherND op is generating a unique name twice which is causing an issue when Casting int32 tensor back to int64 for output. ### Motivation and Context Making sure that the UniqueName function is just called once and reusing it for creating the CastOp.
…microsoft#25986) ### Description * Remove workaround for input1 of FC not supporting uint quantization * Remove workaround for PCQ not being supported on bias of layernorm * Remove workaround for PCQ/uint conv weights unsupported * Fix issue with zero point not being calculated correctly for uint symmetric quantization (microsoft#25823) ### Motivation and Context * Fixes several issues faced by our development team when bringing up a model with PCQ --------- Co-authored-by: Alahari Prudhvi Akhil <prudhvi@qti.qualcomm.com>
**Key changes** This PR integrates KleidiAI SME1 FP32 kernels into the existing kleidiai_sgemm.cpp implementation. Adding SME2 flag in onnxruntime/core/common/cpuid_info.h & onnxruntime/core/common/cpuid_info.cc Previous SME2 kernels integrated were using SME(1) check, this change will correctly distinguish between when SME1 and SME2 kernels are to be used. Bumping KleidiAI version to 1.10.0 **Indicative performance data** Single thread Mac Mini M4 runs on various models using: onnxruntime_perf_test -v -e cpu -I -m times -x 1 -y 1 -r 1 <img width="785" height="400" alt="image" src="https://github.com/user-attachments/assets/37c0b271-14fb-4b76-b2a0-28c5dd9308aa" /> **Next steps** Additional commits to come will address outstanding to-do issues from previous PR linked below: [ KleidiAI SGEMM/IGEMM/Quantized MatMul - Modular MLAS API Changes for KleidiAI microsoft#25187](microsoft#25187) Signed-off-by: Patryk Kaiser <patryk.kaiser@arm.com>
This pull request updates the following GitHub Actions to their latest versions: - `actions/checkout` to `v5` - `actions/setup-python` to `v6` - `actions/setup-node` to `v5` - `actions/setup-java` to `v5` - `actions/setup-dotnet` to `v5` These updates were performed by a script that parsed all YAML workflow files and updated the `uses` statements.
…ft#26015) ### Description <!-- Describe your changes. --> Add opset imports for internal domains when creating a model via Model Editor API. This is consistent with the behavior when loading an ONNX model (see [here](https://github.com/microsoft/onnxruntime/blob/0b96cbef8da027adec09ebabc9d0a260334c1493/onnxruntime/core/graph/model.cc#L224-L255)). ### Motivation and Context <!-- - Why is this change required? What problem does it solve? - If it fixes an open issue, please link to the issue here. --> Required to make the ONNX file valid if the model is saved after optimization. microsoft#25914
This op is used in ORT GenAI and can be decomposed into DequantizeLinear + Gather for WebNN.
ankitm3k
approved these changes
Sep 15, 2025
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
10 participants
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.
Synchronizing intel/onnxruntime ovep-develop branch with latest changes from microsoft/onnxruntime master branch.