Overview

36 Pull requests merged by22 people

• Bump transformers from 4.48.0 to 4.52.1 in /tools/ci_build/requirements/transformers-test

  #25429 mergedJul 18, 2025

• Bump on-headers and compression in /js/react_native/e2e

  #25439 mergedJul 18, 2025

• increase timeout for onnxruntime-ios-packaging-pipeline

  #25438 mergedJul 18, 2025

• Update API ReadOpAttr and CreateOpAttr for string type to unblock EPs to process EPContext node for ep_cache_context with bytes stream

  #25389 mergedJul 18, 2025

• fix shape inference error for ep context nodes

  #25398 mergedJul 18, 2025

• Ovep Feature Rel 1.23

  #25435 mergedJul 18, 2025

• [OVEP] Update OV version to 2025.2.0

  #25436 mergedJul 18, 2025

• Revert "revert qnn sdk version (#25426)"

  #25437 mergedJul 18, 2025

• [QNN EP] Add EP-aware Reshape handler for Transpose optimization.

  #25344 mergedJul 18, 2025

• [CUDA] Support head_sink in flash attention for GQA

  #25432 mergedJul 17, 2025

• Enable free dimension override for graph optimization level 0

  #25425 mergedJul 17, 2025

• [NV RTX EP] Upstream changes from the win-ort

  #25370 mergedJul 17, 2025

• [TRT-EP] Add loadModelProto APIs

  #25409 mergedJul 17, 2025

• add sliding window support for webgpu gqa

  #25372 mergedJul 17, 2025

• Restore ability to handle non-hex string in device discovery vendor/device id.

  #25427 mergedJul 17, 2025

• revert qnn sdk version

  #25426 mergedJul 17, 2025

• Update docker images

  #25418 mergedJul 16, 2025

• [webgpu] Apply template toMatMulNBitsWideTile

  #25353 mergedJul 16, 2025

• Fix Build Error when tensor dumping is enabled

  #25414 mergedJul 16, 2025

• [QNN EP] Gpu backend test framework & test cases.

  #25393 mergedJul 16, 2025

• [webgpu] fix Slice implementation

  #25415 mergedJul 16, 2025

• Fix 2 device discovery issues.

  #25397 mergedJul 16, 2025

• [QNN-EP] Update ScatterND op to reject only QNN-CPU

  #25403 mergedJul 16, 2025

• [QNN-EP] Support GridSample of linear mode for ONNX opset 20+

  #25408 mergedJul 16, 2025

• [CPU] GQA supports attention scores output

  #25319 mergedJul 16, 2025

• Fix SigLIP casual mask bug

  #25360 mergedJul 15, 2025

• Fix some test issues when WebGPU and DML are enabled in the same build

  #25401 mergedJul 15, 2025

• [WebNN] Add rank range validation for rest ops

  #25383 mergedJul 15, 2025

• Bump lintrunner-adapters from 0.12.4 to 0.12.5

  #25380 mergedJul 15, 2025

• Add vendor id to OrtEpFactory and default ORT logger to CreateEpFactories

  #25365 mergedJul 15, 2025

• [QNN EP] Upgrade QNN to 2.36.1

  #25388 mergedJul 14, 2025

• Bump ruff from 0.12.2 to 0.12.3

  #25382 mergedJul 14, 2025

• Bump transformers from 4.48.0 to 4.52.1 in /onnxruntime/python/tools/transformers/models/llama

  #25328 mergedJul 14, 2025

• Fix number of layers in Whisper export

  #25375 mergedJul 14, 2025

• Bump clang-format from 20.1.7 to 20.1.8

  #25381 mergedJul 14, 2025

• [EP ABI] Update to use Node_GetEpName

  #25363 mergedJul 12, 2025

28 Pull requests opened by24 people

• ORT perf test support for plugin EP

  #25374 openedJul 12, 2025

• Optimize layout for SubgroupMatrixLoad on Intel

  #25384 openedJul 14, 2025

• upgrade protobuf in response to security alert

  #25386 openedJul 14, 2025

• [WebGPU EP] allow concat operator to handle large number of inputs

  #25390 openedJul 14, 2025

• [webgpu] use u32 to represent f16 in uniform

  #25391 openedJul 14, 2025

• [webgpu] Optimize FlashAttention for prefill

  #25395 openedJul 15, 2025

• Update fusion_attention to properly convert bfloat16 values

  #25404 openedJul 15, 2025

• 'QnnEpFactory' should provide a fully-qualified path to the backend

  #25407 openedJul 15, 2025

• Fix the is_leaf check in TreeEnsemble

  #25410 openedJul 15, 2025

• [EP ABI] Add documentation for OrtValue and ort_graph_to_proto util

  #25411 openedJul 15, 2025

• add webgpu support for GatherBlockQuantized

  #25413 openedJul 15, 2025

• Subgroup matrix

  #25416 openedJul 16, 2025

• Remove one recursive function in TreeEnsemble

  #25423 openedJul 16, 2025

• [EP ABI] Add Graph_GetModelPath API function

  #25424 openedJul 16, 2025

• [CANN]Fix issue with negative dynamic tensor shape

  #25431 openedJul 17, 2025

• [WebNN] Fix some spelling and naming issues

  #25433 openedJul 17, 2025

• [webgpu] support And operator

  #25440 openedJul 18, 2025

• [webgpu] support float16 type for Einsum operator

  #25443 openedJul 18, 2025

• [QNN EP] Enable Conv Op with "auto_pad" param set as VALID

  #25444 openedJul 18, 2025

• [NV RTX EP] Set Compute Capability only on Turing architecture

  #25446 openedJul 18, 2025

• [VitisAI] Remove 4k alignment from preferred allocator

  #25447 openedJul 18, 2025

• [VitisAI] Upstream changes from win-ort

  #25448 openedJul 18, 2025

• Iraut/vendor id impl

  #25449 openedJul 18, 2025

• Update .config/1espt/PipelineAutobaseliningConfig.yml

  #25450 openedJul 18, 2025

• Remove training packages from onnxruntime-ios-packaging-pipeline

  #25451 openedJul 18, 2025

• Enable TSA for nuget packaging pipelines

  #25452 openedJul 18, 2025

• [QNN EP] Minor fix to enable MatMulAddFusion

  #25453 openedJul 18, 2025

• [EP ABI] Signatures for compatibility info methods

  #25454 openedJul 18, 2025

11 Issues closed by3 people

• [Performance]

  #24787 closedJul 18, 2025

• ONNX model loading crashes on MacOS in protobuf function

  #25420 closedJul 16, 2025

• [Build] compilation error: invalid instruction mnemonic 'vcvtneeph2ps'

  #22519 closedJul 15, 2025

• [Build] Unable to build ONNX Runtime 1.22 due to dependency update

  #25098 closedJul 15, 2025

• [Build] Unable to build due to eigen dependency

  #25406 closedJul 15, 2025

• why not support ceil() in version 1.12.1

  #24674 closedJul 15, 2025

• [Build] Cannot cross platform build for v3.20 alpine arm64

  #24788 closedJul 15, 2025

• [Web] Unable to build using WebGPU - `error: handleI64Signatures: signature too long for emwgpuWaitAny`

  #25359 closedJul 14, 2025

• [Build] Fail to pass AutoEpSelection and OrtEpLibrary tests in Windows x64 QNN build

  #24676 closedJul 14, 2025

• [Documentation] outdated documents about cuda version and onnxruntime

  #24759 closedJul 14, 2025

• pip install --no-cache-dir onnxruntime-gpu --no-deps fails to install

  #25288 closedJul 13, 2025

16 Issues opened by16 people

• int4 quantized model can't run with DML provider

  #25445 openedJul 18, 2025

• [Performance] NV TRT RTX provider performance slower than TensorRT on RTX 4000 Ada

  #25442 openedJul 18, 2025

• [Mobile] QNN graph execute error. Error code: 6001

  #25422 openedJul 16, 2025

• [Feature Request] Support BERT models In CoreML execution provider

  #25421 openedJul 16, 2025

• The resize operator's behavior may use Round instead of Floor

  #25417 openedJul 16, 2025

• Support output QK symbolic shape inference

  #25412 openedJul 15, 2025

• LogSoftmax produces different results when opset=11 or 13

  #25402 openedJul 15, 2025

• TreeEnsemble dies due to memory/segmentation errors with large membership values

  #25400 openedJul 15, 2025

• [Build] libonnxruntime_providers_cuda.so built from soure (v1.22) is too large (1.4GB)

  #25399 openedJul 15, 2025

• [Performance] Update ONNX runtime to increase performance on CoreML

  #25396 openedJul 15, 2025

• Can model Optimizations support skip custom operator?

  #25394 openedJul 15, 2025

• [Feature Request] [C#] Add support for System.Half for input/output tensors

  #25392 openedJul 14, 2025

• [Feature Request] Injecting GPU memory in CUDA/TensorRT EPs

  #25385 openedJul 14, 2025

• Invalid MIGraphX EP option: migraphx_load_compiled_path

  #25379 openedJul 13, 2025

• [Performance] DequantizeLinear performance with `uint8` input compared to `int4`/`uint4`. Is casting between uint8->int4/uint4 possible?

  #25378 openedJul 13, 2025

• [Build] Can't build CUDA on Windows

  #25377 openedJul 13, 2025

48 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Attention Operator (CPU)

  #25156 commented onJul 18, 2025 • 42 new comments

• [MIGraphx EP] Sync AMD changes upstream

  #25338 commented onJul 18, 2025 • 35 new comments

• Plugin EP data transfer and Stream support.

  #25254 commented onJul 18, 2025 • 20 new comments

• [EP ABI] Get EP compiled model compatibility

  #25331 commented onJul 18, 2025 • 10 new comments

• Convert Initializers to OrtValues Phase 2

  #25320 commented onJul 18, 2025 • 8 new comments

• [Mlas] optimize MlasConv using thread partition opt

  #25255 commented onJul 15, 2025 • 4 new comments

• Enable CUDA Graph in nv_tensorrt_rtx EP

  #25368 commented onJul 18, 2025 • 2 new comments

• Iraut/update nv trt rtx ep doc

  #25321 commented onJul 15, 2025 • 2 new comments

• Bump protobuf from 3.20.2 to 4.25.8 in /onnxruntime/python/tools/transformers/models/llama

  #25085 commented onJul 18, 2025 • 0 new comments

• Bump protobuf from 3.20.2 to 4.25.8 in /onnxruntime/python/tools/transformers/models/whisper

  #25086 commented onJul 18, 2025 • 0 new comments

• Bump protobuf from 3.20.3 to 4.25.8 in /onnxruntime/python/tools/transformers/models/stable_diffusion/requirements

  #25087 commented onJul 18, 2025 • 0 new comments

• Bump protobuf from 4.21.12 to 4.25.8 in /tools/ci_build/github/linux/docker/inference/aarch64/python/cpu/scripts

  #25088 commented onJul 18, 2025 • 0 new comments

• Update index.md

  #25119 commented onJul 17, 2025 • 0 new comments

• KleidiAI SGEMM/IGEMM/Quantized MatMul - Modular MLAS API Changes for KleidiAI

  #25187 commented onJul 16, 2025 • 0 new comments

• [WIP] Add some device discovery support for non-Windows platforms

  #25228 commented onJul 15, 2025 • 0 new comments

• [ARM CPU] SVE support for Elementwise kernels

  #25238 commented onJul 14, 2025 • 0 new comments

• Upgrade xnnpack to latest

  #25275 commented onJul 14, 2025 • 0 new comments

• FIX: dxcore include when compiling with older Windows SDK

  #25297 commented onJul 13, 2025 • 0 new comments

• Bump transformers from 4.50.0 to 4.51.0 in /onnxruntime/python/tools/transformers/models/stable_diffusion/requirements

  #25322 commented onJul 18, 2025 • 0 new comments

• [NvTensorRTRTX EP]Disable Fast GELU operator in base model used for NV EP Unit Tests

  #25323 commented onJul 14, 2025 • 0 new comments

• Update python bindings to be able to use a shared allocator and/or IDataTransfer registered by a plugin EP in the Environment

  #25346 commented onJul 14, 2025 • 0 new comments

• Support read-only allocator for use with initializers

  #25348 commented onJul 14, 2025 • 0 new comments

• Add patch for WebGPU on Android to handle fp16 in uniforms

  #25349 commented onJul 17, 2025 • 0 new comments

• [webgpu] Enable per-run control for graph capture

  #25367 commented onJul 14, 2025 • 0 new comments

• [Build] Failed to build onnxruntime java api 1.20.0 on windows (C2039: 'system_clock': is not a member of 'std::chrono')

  #24622 commented onJul 12, 2025 • 0 new comments

• Non-zero status code returned while running LSTM node

  #10768 commented onJul 12, 2025 • 0 new comments

• onnxruntime outputs different results for different opset versions

  #25050 commented onJul 13, 2025 • 0 new comments

• [Build] ORT can't build with cuda 12.9

  #24731 commented onJul 13, 2025 • 0 new comments

• SafeIntOnOverflow() Integer overflow error when running inference in an ASGI server

  #12288 commented onJul 14, 2025 • 0 new comments

• onnxruntime-directml import interference with sklearn

  #21724 commented onJul 14, 2025 • 0 new comments

• [Delivery] Win ARM64 wheels + QNN

  #19162 commented onJul 14, 2025 • 0 new comments

• Multi-threaded GPU inferencing failing with whisper-small: Non-zero status code returned while running DecoderMaskedMultiHeadAttention node

  #21413 commented onJul 15, 2025 • 0 new comments

• OnnxRuntimeGenAIException: CUDA execution provider is not enabled in this build.

  #23715 commented onJul 15, 2025 • 0 new comments

• dockerfile with different stage

  #17812 commented onJul 16, 2025 • 0 new comments

• [Mobile] Android Native crash in [split_config.armeabi_v7a.apk!libonnxruntime.so] OrtSessionOptionsAppendExecutionProvider_Nnapi

  #25138 commented onJul 16, 2025 • 0 new comments

• Persistent Crashes on Android/armeabi-v7a

  #25097 commented onJul 16, 2025 • 0 new comments

• How can I output every node's output shape when infer with onnx models with lots of if branches

  #25052 commented onJul 16, 2025 • 0 new comments

• [Mobile] MatMulNbits Q8 Errors out on Android

  #24769 commented onJul 16, 2025 • 0 new comments

• [Build] Build fails: 'error : no operator "+=" matches these operands' with nv_bfloat16

  #25162 commented onJul 16, 2025 • 0 new comments

• [Mobile] TypeError: A bool tensor's data must be type of function Uint8Array() { [native code] }

  #25294 commented onJul 16, 2025 • 0 new comments

• [Web] Fail to link static Wasm library with WebNN EP support

  #24936 commented onJul 17, 2025 • 0 new comments

• [Performance] How does onnxruntime run in parallel mode?

  #21259 commented onJul 17, 2025 • 0 new comments

• Static quantize self-attention module not work

  #17278 commented onJul 17, 2025 • 0 new comments

• ORT 1.22.0 fails assertion on python import on aarch64, but not x86_64

  #25103 commented onJul 18, 2025 • 0 new comments

• [Web] Cannot import from web worker

  #25096 commented onJul 18, 2025 • 0 new comments

• Failed to load library libonnxruntime_providers_cuda.so I am getting the following erro

  #19616 commented onJul 18, 2025 • 0 new comments

• [TRT RTX EP] Implement GetEPContextNodes()

  #24901 commented onJul 18, 2025 • 0 new comments

• [MIGRAPHX] Add ORT generic interface build support for MigraphX

  #25004 commented onJul 14, 2025 • 0 new comments