Issue Labeler

labeler.yml

1960 workflow runs

[Performance] CUDA Conv: cuDNN frontend >= 1.24 calls cudaGetDeviceProperties on every plan rebuild (~1 ms per Conv node per input-shape change)
completed successfully:
Issue Labeler#10781:Issue#33769openedbysilviot
Oct 30, 202615s
Float8 QDQ models fail with INVALID_GRAPH on CPU EP at ORT_ENABLE_EXTENDED and above
completed successfully:
Issue Labeler#10780:Issue#33761openedbysisleyli
Oct 30, 202617s
Plugin-EP partition names collide under ep.share_ep_contexts: byte-loaded models share one identity, and ep.context_node_name_prefix is ignored
completed successfully:
Issue Labeler#10779:Issue#33758openedbyadrastogi
Oct 30, 202630s
[Web] WebGPU SkipLayerNormalization / SkipSimplifiedLayerNormalization return zeros past maxStorageBufferBindingSize
completed successfully:
Issue Labeler#10778:Issue#33755openedbymohdmahmodi
Oct 30, 20266m 13s
[Web] [WebGPU] Single-pass prefill of ~900 tokens crashes Safari's GPU process on iOS 27 (iPhone) — please split large workloads into smaller submissions / dispatches
completed successfully:
Issue Labeler#10777:Issue#33754openedbydaveycodez
Oct 29, 202617m 29s
[Web] [WebGPU] Single-pass prefill of ~900 tokens crashes Safari's GPU process on iOS 27 (iPhone) — please split large workloads into smaller submissions / dispatches
completed successfully:
Issue Labeler#10776:Issue#33753openedbydaveycodez
Oct 29, 202617m 41s
[Feature Request] [Web] Expose kvCacheQuantizationBits in the onnxruntime-web WebGPU EP options
completed successfully:
Issue Labeler#10775:Issue#33752openedbydaveycodez
Oct 29, 202620m 5s
[Web] WebGPU bucket buffer cache: 4 MiB → 8 MiB step and low large-bucket limits cause a memory jump and allocation churn in long-context LLM decoding; limits not configurable
completed successfully:
Issue Labeler#10774:Issue#33751openedbydaveycodez
Oct 29, 20264m 16s
[Web] WebGPU EP creates a new GPUBindGroup for every dispatch (~317 per decoded token); bind groups are costly in WebKit (memory held until GC)
completed successfully:
Issue Labeler#10773:Issue#33750openedbydaveycodez
Oct 29, 20264m 17s
[Feature Request] EP-agnostic callbacks for persisting EP runtime/compilation caches
completed successfully:
Issue Labeler#10772:Issue#33748openedbyadrastogi
Oct 29, 202617s
[Web] Firefox throws URL constructor: null is not a valid URL when opening Vite single-file build via file://
completed successfully:
Issue Labeler#10771:Issue#33735openedbyraimundoprez
Oct 29, 20263m 47s
[Bug][DirectML EP] Reshape with allowzero=1 and a -1 dimension misinfers the output shape and crashes
completed successfully:
Issue Labeler#10770:Issue#33720openedbyPulsarUnderscore
Oct 29, 20263m 8s
[Build] Both 1.31.0 and 1.31.1 failed to Build Windows msvc2019 x64 Release with error C2440
completed successfully:
Issue Labeler#10769:Issue#33636editedbyJingDict
Oct 29, 20265m 42s
[CPU EP] MatMulNBits (accuracy_level=4) produces wrong results on x86-64 AVX2 when loaded from session.save_external_prepacked_constant_initializers output (1.31.1)
completed successfully:
Issue Labeler#10768:Issue#33677openedbythiennq
Oct 28, 20268m 11s
[Web/WebGPU] 1.31: a session created on a page-supplied GPUDevice cannot bind gpu-buffer inputs
completed successfully:
Issue Labeler#10767:Issue#33642openedbyczoli1976
Oct 27, 20263m 35s
[Build] CUDA EP fails to load in training-enabled builds since #29252: backward cuDNN calls bypass the dynamic loader
completed successfully:
Issue Labeler#10766:Issue#33638openedbyreposi-thor-y
Oct 27, 20263m 34s
[Build] Both 1.31.0 and 1.31.1 failed to Build Windows msvc2019 x64 Release with error C2440
completed successfully:
Issue Labeler#10765:Issue#33636openedbyJingDict
Oct 27, 20264m 41s
[Build] cannot build with --disable_contrib_ops on Windows CUDA
completed successfully:
Issue Labeler#10764:Issue#33602editedbysevenc-nanashi
Oct 27, 20263m 38s
[Plugin EP] Provider-bridge probe unloads the EP DLL, so RegisterExecutionProviderLibrary loads and initializes it twice
completed successfully:
Issue Labeler#10763:Issue#33603editedbymingyueliuh
Oct 26, 20263m 41s
[Plugin EP] Provider-bridge probe unloads the EP DLL, so RegisterExecutionProviderLibrary loads and initializes it twice
completed successfully:
Issue Labeler#10762:Issue#33603editedbymingyueliuh
Oct 26, 20263m 36s
[Plugin EP] Provider-bridge probe unloads the EP DLL, so RegisterExecutionProviderLibrary loads and initializes it twice
completed successfully:
Issue Labeler#10761:Issue#33603openedbymingyueliuh
Oct 26, 20263m 43s
[Build] cannot build with --disable_contrib_ops on Windows CUDA
completed successfully:
Issue Labeler#10760:Issue#33602editedbysevenc-nanashi
Oct 26, 20263m 41s
[Build] cannot build with --disable_contrib_ops on Windows CUDA
completed successfully:
Issue Labeler#10759:Issue#33602openedbysevenc-nanashi
Oct 26, 20263m 46s
[CPU] ReduceSum loses int64 precision even when the result does not overflow
completed successfully:
Issue Labeler#10758:Issue#33579openedbygalleonli
Oct 25, 20263m 38s
MatMulNBits benchmark generates IR12 models rejected by ORT 1.23.2
completed successfully:
Issue Labeler#10757:Issue#33576openedbyarcusbuilds
Oct 24, 20263m 39s

Navigation Menu

Issue Labeler

1960 workflow runs

[Performance] CUDA Conv: cuDNN frontend >= 1.24 calls cudaGetDeviceProperties on every plan rebuild (~1 ms per Conv node per input-shape change)

Float8 QDQ models fail with INVALID_GRAPH on CPU EP at ORT_ENABLE_EXTENDED and above

Plugin-EP partition names collide under `ep.share_ep_contexts`: byte-loaded models share one identity, and `ep.context_node_name_prefix` is ignored

[Web] WebGPU SkipLayerNormalization / SkipSimplifiedLayerNormalization return zeros past maxStorageBufferBindingSize

[Web] [WebGPU] Single-pass prefill of ~900 tokens crashes Safari's GPU process on iOS 27 (iPhone) — please split large workloads into smaller submissions / dispatches

[Web] [WebGPU] Single-pass prefill of ~900 tokens crashes Safari's GPU process on iOS 27 (iPhone) — please split large workloads into smaller submissions / dispatches

[Feature Request] [Web] Expose kvCacheQuantizationBits in the onnxruntime-web WebGPU EP options

[Web] WebGPU bucket buffer cache: 4 MiB → 8 MiB step and low large-bucket limits cause a memory jump and allocation churn in long-context LLM decoding; limits not configurable

[Web] WebGPU EP creates a new GPUBindGroup for every dispatch (~317 per decoded token); bind groups are costly in WebKit (memory held until GC)

[Feature Request] EP-agnostic callbacks for persisting EP runtime/compilation caches

[Web] Firefox throws URL constructor: null is not a valid URL when opening Vite single-file build via file://

[Bug][DirectML EP] Reshape with allowzero=1 and a -1 dimension misinfers the output shape and crashes

[Build] Both 1.31.0 and 1.31.1 failed to Build Windows msvc2019 x64 Release with error C2440

[CPU EP] MatMulNBits (accuracy_level=4) produces wrong results on x86-64 AVX2 when loaded from `session.save_external_prepacked_constant_initializers` output (1.31.1)

[Web/WebGPU] 1.31: a session created on a page-supplied GPUDevice cannot bind gpu-buffer inputs

[Build] CUDA EP fails to load in training-enabled builds since #29252: backward cuDNN calls bypass the dynamic loader

[Build] Both 1.31.0 and 1.31.1 failed to Build Windows msvc2019 x64 Release with error C2440

[Build] cannot build with --disable_contrib_ops on Windows CUDA

[Plugin EP] Provider-bridge probe unloads the EP DLL, so RegisterExecutionProviderLibrary loads and initializes it twice

[Plugin EP] Provider-bridge probe unloads the EP DLL, so RegisterExecutionProviderLibrary loads and initializes it twice

[Plugin EP] Provider-bridge probe unloads the EP DLL, so RegisterExecutionProviderLibrary loads and initializes it twice

[Build] cannot build with --disable_contrib_ops on Windows CUDA

[Build] cannot build with --disable_contrib_ops on Windows CUDA

[CPU] ReduceSum loses int64 precision even when the result does not overflow

MatMulNBits benchmark generates IR12 models rejected by ORT 1.23.2

Workflow runs for Issue Labeler

1960 workflow runs