1Release published by1person

• v1.19.2 ONNX Runtime v1.19.2

  published Sep 4, 2024

47 Pull requests merged by 25people

• Move Gelu and LayerNorm fusion to L1 optimization

  #21332mergedSep 9, 2024

• Add parameter for flexdonwload

  #22009mergedSep 8, 2024

• [WebNN EP] Remove workaround for CPU op supported list

  #21962mergedSep 7, 2024

• Use output variable from InstallAppleProvisioningProfile task to set provisioning profile UUID.

  #22018mergedSep 7, 2024

• [VitisAI] Add processing for sessionOptions.AppendExecutionProvider( "VitisAI", options)

  #21839mergedSep 6, 2024

• NimbleEdge blog update

  #22017mergedSep 6, 2024

• near-zero negative values must convert to 0 not NAN

  #18473mergedSep 6, 2024

• remove unused and confusing float16 constants

  #21999mergedSep 6, 2024

• Fix typo in coreml_supported_mlprogram_ops.md

  #22004mergedSep 6, 2024

• Update Android NDK version to 27.0.12077973.

  #21989mergedSep 6, 2024

• [TransposeOptimizer] Support Unsqueeze/Transpose of input consumed by per-axis DQ

  #21821mergedSep 6, 2024

• [WebNN EP] Use identity for one input of Max/Min

  #21974mergedSep 5, 2024

• Update CoreML supported ops lists

  #21627mergedSep 5, 2024

• Add better native nuget package readme

  #21889mergedSep 5, 2024

• [CUDA] Update Dockerfile.cuda with cuda 12.5.1 and cudnn 9

  #21987mergedSep 5, 2024

• [Fuzzer] Add fuzzer support for linux

  #21996mergedSep 5, 2024

• [VitisAI] remove unused header

  #21890mergedSep 5, 2024

• Uncomment line in OVEP that was commented out in error

  #21973mergedSep 5, 2024

• Fix DML packaging CIs

  #21997mergedSep 5, 2024

• complete the GetSubGraph

  #21998mergedSep 5, 2024

• Fix C# doc generation workflow

  #21988mergedSep 5, 2024

• fix one build warning in MSVC

  #21983mergedSep 5, 2024

• [js/webgpu] Optimize grouped conv

  #21892mergedSep 5, 2024

• Add packaging version constraint.

  #21814mergedSep 4, 2024

• Sets enable_windows_arm64ec_qnn to false in training CI

  #21981mergedSep 4, 2024

• Update C# test projects

  #21631mergedSep 4, 2024

• Update C# E2E project's test package versions

  #21975mergedSep 4, 2024

• Rename ios_packaging.requirements.txt to ios_packaging/requirements.txt

  #21936mergedSep 4, 2024

• [js/webgpu] Optimize transpose

  #21964mergedSep 4, 2024

• Enable QNN weight sharing

  #21077mergedSep 4, 2024

• Update documentation for ovep rel-5.4

  #21909mergedSep 4, 2024

• [VitisAI] add registered custom op for perf test

  #21336mergedSep 4, 2024

• [VitisAI] Bug fixes in model_clone

  #21950mergedSep 4, 2024

• Improve stability of Android ReactNative E2E test

  #21969mergedSep 4, 2024

• Update vsinpu ep cross-compiling patch

  #21963mergedSep 4, 2024

• [VitisAI] Fix model path

  #21911mergedSep 4, 2024

• refactor: extract shared util function ComputeBroadcastOutputShape

  #21940mergedSep 4, 2024

• compare the content of WEBGPU_BUFFER, not the address

  #21967mergedSep 3, 2024

• reflect change in WebGpuProviderFactoryCreator::Create signature

  #21971mergedSep 3, 2024

• Remove unused find_cudnn_supported_cuda_versions

  #21620mergedSep 3, 2024

• Memory Optimization for Compilation in OVEP

  #21872mergedSep 3, 2024

• Add Expand op

  #21933mergedSep 3, 2024

• Enable xnnpack ep works in current windows xnn ci

  #21951mergedSep 3, 2024

• Fix copying ORT dylib into wheel on macOS

  #21931mergedSep 3, 2024

• revert forceinline for MakeString

  #21943mergedSep 3, 2024

• Fix C# warnings.

  #21913mergedSep 3, 2024

• Add dependency dawn into deps.txt

  #21910mergedSep 2, 2024

20 Pull requests opened by16people

• [Quantization] Apply workaround for crash when using histogram-based calibrators

  #21972openedSep 3, 2024

• fix more dml warnings

  #21980openedSep 4, 2024

• configure logger for onnxruntime-node package

  #21982openedSep 4, 2024

• Implementation of AVX-VNNI-INT8 dot product instructions into MLAS GEMM

  #21984openedSep 4, 2024

• adds support for Uint8ClampedArray

  #21985openedSep 4, 2024

• [webgpu-native] Add transpose op

  #21986openedSep 4, 2024

• Add FastGelu op

  #21991openedSep 5, 2024

• [ROCm] fix rocm-6.2 build issues

  #21993openedSep 5, 2024

• [JS/WebGPU] Upgrade emsdk to 3.1.61

  #21994openedSep 5, 2024

• [WIP][js/webgpu] Don't do layout conversion for InstanceNormalization

  #21995openedSep 5, 2024

• Enable Pad->Conv(no pads) fusion

  #22001openedSep 5, 2024

• Fixed many (not all) accessibility issues.

  #22002openedSep 5, 2024

• Decrease API docs artifact retention days

  #22003openedSep 5, 2024

• [Draft] Run tests on old pending allocator change

  #22008openedSep 6, 2024

• Upgrade XNNPACK to latest version

  #22012openedSep 6, 2024

• [webgpu-native] Add where op

  #22014openedSep 6, 2024

• Prevent int32 quantized bias from clipping by adjusting the weight's scale

  #22020openedSep 7, 2024

• [WebNN EP] Use opSupportLimits to dynamically check data type support

  #22025openedSep 9, 2024

• Add fusions for re-designed Phi-3 vision and Phi-3.5 vision ONNX models

  #22026openedSep 9, 2024

• Ovep release lnl 1.2.1

  #22027openedSep 9, 2024

6Issues closed by5people

• [Feature Request] Move graph compilation behind higher transformers (graph optimization)

  #20915closedSep 9, 2024

• E:onnxruntime:Default, provider_bridge_ort.cc:1992 onnxruntime::TryGetProviderInfo_CUDA

  #22019closedSep 6, 2024

• [Mobile] Test Application Reveals Multiple Failures in QnnHTPBackendTests Suite on Device

  #21887closedSep 5, 2024

• [Build] 1.19.0 still depends on CUDA 11.x

  #21965closedSep 4, 2024

• Llama2 RMS Norm: SimplifiedLayerNormalization

  #21925closedSep 4, 2024

• Llama2 RMS Norm: SimplifiedLayerNormalization

  #21924closedSep 3, 2024

21Issues opened by21people

• [Web] onnxruntime-gpu(1.18.0) can not be install

  #22028openedSep 9, 2024

• [Inference & Training] My Onnxruntime isnt detecting cuda even after all paths are perfectly given with compatible softwares

  #22016openedSep 6, 2024

• [Build] onnxruntime-openvino library does not have Python 3.12 support

  #22015openedSep 6, 2024

• onnxruntime c++ logging crash

  #22013openedSep 6, 2024

• [Web] no available backend found [wasm] when importing `onnxruntime-web/wasm`

  #22010openedSep 6, 2024

• Context leak detected with CoreMLExecutionProvider

  #22007openedSep 6, 2024

• Meet error [No module named '_kernel_explorer'] when use triton kernel

  #22006openedSep 6, 2024

• [Web] __turbopack_resolve_absolute_path__ is not a function

  #22005openedSep 6, 2024

• CUDA does not load on Windows

  #22000openedSep 5, 2024

• Extra Memory Usage due to 'CreateSessionFromArray' copying the model again

  #21992openedSep 5, 2024

• [Performance] CUDAExecutionProvider without RoiAlign (opset 16 version)

  #21990openedSep 5, 2024

• [Performance] Increasing Memory Usage during INT8 Quantization with ONNX Runtime tools

  #21979openedSep 4, 2024

• [Build] compiling the WASM in Firefox takes ~10 minutes and 4GB of ram

  #21978openedSep 4, 2024

• Has there been a 1.19.2 release (tag was set)

  #21977openedSep 4, 2024

• [Build] CMake Fails to Find Non-LTS Abseil Library (Download Triggered)

  #21976openedSep 4, 2024

• [Web] Uncaught WebGPU validation error on Snapdragon SM8450 but works on SM8250

  #21970openedSep 3, 2024

• [Web] BiRefNet_T not working on webgpu

  #21968openedSep 3, 2024

• [CUDA][Performance] Inference time greatly variates during session run

  #21966openedSep 3, 2024

• [Mobile] IOS library crashes in Release configuration

  #21960openedSep 2, 2024

• Use AppendExecutionProvider_Dnnl api to add onednn EP, No success.

  #21958openedSep 2, 2024

• 1.19: Clip operator with type FLOAT16 defaults to min or max value 0.0 if not explicitly given, breaking many models using FLOAT16

  #21957openedSep 2, 2024

50 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.

• [WebNN EP] Support GRU operator

  #20405 commented onSep 9, 2024• 24 new comments

• Refactor CoreMLExecution to C++ bridge class

  #21857 commented onSep 8, 2024• 7 new comments

• Adding ONNX Runtime C-API for WebGPU EP

  #21838 commented onSep 5, 2024• 5 new comments

• Using wostringstream only on Windows

  #21938 commented onSep 6, 2024• 4 new comments

• [WIP] WebGPU EP [skip ci]

  #21904 commented onSep 9, 2024• 2 new comments

• [Web] WebGPU and WASM Backends Unavailable within Service Worker

  #20876 commented onSep 9, 2024• 0 new comments

• Update pool to MacOS-13

  #17361 commented onSep 5, 2024• 0 new comments

• Update C++ to standard 20 for Windows

  #17706 commented onSep 6, 2024• 0 new comments

• Enable AVX NE CONVERT for FP16 to FP32 cast

  #21183 commented onSep 6, 2024• 0 new comments

• Create CMake option `onnxruntime_USE_VCPKG`

  #21348 commented onSep 6, 2024• 0 new comments

• [WIP] Out-Tree EP feature

  #21450 commented onSep 6, 2024• 0 new comments

• Add Interactive Decoding support in GQA

  #21523 commented onSep 6, 2024• 0 new comments

• ThreadPool: Spend less time busy waiting.

  #21545 commented onSep 5, 2024• 0 new comments

• Make argmin/armax support identical data types and add int64 support

  #21641 commented onSep 4, 2024• 0 new comments

• Add option for compiler flags.

  #21721 commented onSep 3, 2024• 0 new comments

• ConvTranpose using CUDNN Frontend with NHWC support

  #21752 commented onSep 9, 2024• 0 new comments

• [java] Fix for OnnxTensor creation when passing in a ByteBuffer containing elements of a different type

  #21774 commented onSep 9, 2024• 0 new comments

• Matmul_nbits kernel for mlas sqnbits to support Fp16 inputs

  #21807 commented onSep 5, 2024• 0 new comments

• Genai page addition

  #21862 commented onSep 5, 2024• 0 new comments

• Integrate onnx 1.17.0

  #21897 commented onSep 5, 2024• 0 new comments

• [VitisAI] Translate all session configs into provider options with prefix

  #21907 commented onSep 4, 2024• 0 new comments

• fmha slide window

  #21926 commented onSep 5, 2024• 0 new comments

• Set Transpose Attribute instead for manipulating MatMul Strides

  #21927 commented onSep 4, 2024• 0 new comments

• [CUDA] upgrade cutlass to 3.5.1

  #21939 commented onSep 6, 2024• 0 new comments

• [JS/WebGPU] Fixed bugs in inputs validation of Resize

  #21955 commented onSep 3, 2024• 0 new comments

• CUDA_PATH is set but CUDA wasnt able to be loaded

  #21527 commented onSep 2, 2024• 0 new comments

• New restricted asymmetric quantization mode in QDQ mode with zero_point restricted to either 128 or 0

  #21398 commented onSep 2, 2024• 0 new comments

• CreateSessionFromArray doesn't work

  #21946 commented onSep 3, 2024• 0 new comments

• TensorrtExecutionProvider slower than CUDAExecutionProvider: Faster-rcnn [Performance]

  #17434 commented onSep 3, 2024• 0 new comments

• Why C++ cannot modify the enable_mem_reuse option in Ort::SessionOptions...

  #21942 commented onSep 3, 2024• 0 new comments

• Importing onnxruntime on AWS Lambdas with ARM64 processor causes crash

  #10038 commented onSep 4, 2024• 0 new comments

• CPU cores and threads control

  #8193 commented onSep 4, 2024• 0 new comments

• Python ORT latest is available in 1.19.0 whereas NuGet package is 1.19.1

  #21953 commented onSep 4, 2024• 0 new comments

• [Discussion] ORT GPU binaries do not contain DML

  #20638 commented onSep 5, 2024• 0 new comments

• [Web] Memory access out of bounds / alignment fault

  #21355 commented onSep 5, 2024• 0 new comments

• [Web] How to free webgpu gpu mem in onnxruntime web

  #21574 commented onSep 5, 2024• 0 new comments

• DirectML returning empty result with ObjectDetection (Mobilinet V2 FPN Keras)

  #20386 commented onSep 5, 2024• 0 new comments

• [Mobile] Will provide supported 16KB-page-size prebuilt Android.so in the future?

  #21837 commented onSep 5, 2024• 0 new comments

• Failed to allocated memory for requested buffer of size X

  #20038 commented onSep 6, 2024• 0 new comments

• [Build] CUDA Illegal Memory Access error when using a custom Triton kernel

  #20885 commented onSep 6, 2024• 0 new comments

• Java GPU dependency of ONNX Runtime version 1.18 only support CUDA 12?

  #21651 commented onSep 6, 2024• 0 new comments

• Android build: Execution failed for task ':app:mergeExtDexDebug'.

  #21494 commented onSep 6, 2024• 0 new comments

• DirectML error: The parameter is incorrect with KBNet S

  #21583 commented onSep 7, 2024• 0 new comments

• [Mobile] [react-native] [android] Tensor.fromImage error ReferenceError: Property 'document' doesn't exist

  #17752 commented onSep 7, 2024• 0 new comments

• [cuda ep] Squeeze node fails when axes is not provided

  #21661 commented onSep 7, 2024• 0 new comments

• [Performance] Mapfile support for certain external data files is not working

  #21195 commented onSep 7, 2024• 0 new comments

• how to release gpu memory when keep onnxruntime session around.

  #9509 commented onSep 8, 2024• 0 new comments

• [Performance] Inference time discrepancy when using TorchScript vs ONNX exported model

  #21689 commented onSep 8, 2024• 0 new comments

• [Feature Request] Support for Florence-2 model family

  #21118 commented onSep 8, 2024• 0 new comments

• [Web] Wav2vec2 slower on WebGPU than WASM

  #21618 commented onSep 9, 2024• 0 new comments