All pull requests

Pull requests

• [WebGPU] Allow exact runtime compatibility exceptions

  #32787·Xiaofei Han (xiaofeihan1) opened Sep 24, 2026Contributor·

  ·
  2

• [CPU] Fix INT2 QuantizeLinear per-axis output on the last axis

  #32786·TANGBUDU opened Sep 24, 2026·

  ·
  2

• Fix WebGPU Concat with many inputs

  #32783·tianleiwu with CopilotAI opened Sep 24, 2026·

  ·
  1

  1

  

• [WIP] Fix CPUExecutionProvider output differences for FusedConv model

  #32782·tianleiwu with CopilotAI opened Sep 24, 2026·

  ·
  1

  1

  

• Default Java GPU packaging to CUDA 13

  #32781·tianleiwu with CopilotAI opened Sep 24, 2026·

  ·
  1

  1

  

• Preserve EP registration End events for returned errors without telemetry

  #32780·bmehta001 opened Sep 24, 2026Contributor·

  ·
  1

• Add ability for app to pin ORT version when consuming onnxruntime-react-native npm package

  #32779·Scott McKay (skottmckay) opened Sep 24, 2026Contributor·

  ·
  1

• [Plan] Multi-device EP support

  #32778·Scott McKay (skottmckay) opened Sep 24, 2026Contributor·

  ·

• fix: name the activation the WebGPU Conv kernel cannot fuse (#32716)

  #32775·Yuri Khrustalev (ykhrustalev) opened Sep 24, 2026Contributor·

  ·
  2

• Fold Transpose on constant initializer instead of pushing it

  #32770·Yuduo Wu (1duo) opened Sep 23, 2026Contributor·

  ·
  4

• Test Windows 2025 migration

  #32769·eserscor opened Sep 23, 2026Contributor·

  ·

• Split webgpu build into build and test. Reduce total runtime ~45 mins.

  #32768·eserscor opened Sep 23, 2026Contributor·

  ·
  5

• [WebGPU EP] adds state update capacity support for GatedDeltaNet

  #32767·Prathik Rao (prathikr) opened Sep 23, 2026Contributor·

  ·
  7

• Enable non-causal WebGPU PagedAttention for full-context blocks

  #32763·aciddelgado opened Sep 23, 2026Contributor·

  ·
  7

• [WIP] Fix failing GitHub Actions job Build Linux x64 Release

  #32762·mirounga with CopilotAI opened Sep 23, 2026·

  ·
  1

  

• Add packed INT2 and mixed-width CUDA QMoE decode

  #32761·David Fan (jiafatom) opened Sep 23, 2026Contributor·

  ·
  5

• cmake: install onnxruntime_perf_test binary

  #32759·pratikpachange97 opened Sep 23, 2026·

  ·
  2

• [MLAS] Add AArch64 SVE FP16 GEMM kernel and driver

  #32756·Nikhil Sharma (nikhilfujitsu) opened Sep 23, 2026Contributor·

  ·
  7

• [MLAS] Fix fp32 overflow/underflow in ARM64 NEON QLinearMul scale ratio

  #32754·kjg0724 (kjg0724) opened Sep 23, 2026Contributor·

  ·
  4

• Validate sparse initializer value cardinality

  #32753·Akshay Sonawane (apsonawane) opened Sep 23, 2026Contributor·

  ·
  7

• Upgrade transformers to 5.3.0

  #32752·danielsongmicrosoft opened Sep 22, 2026Contributor·

  ·
  1

• Add sliding window AVX-512 NCHWc depthwise kernel (bitwise identical)

  #32749·mirounga opened Sep 22, 2026Contributor·

  ·
  10

• Synchronize telemetry state capture

  #32748·Akshay Sonawane (apsonawane) opened Sep 22, 2026Contributor·

  ·
  3

• Validate DML upload heap allocations

  #32745·Akshay Sonawane (apsonawane) opened Sep 22, 2026Contributor·

  ·
  5

• Expand CUDA QMoE GEMV rows on SM120+

  #32741·Baiju Meswani (baijumeswani) opened Sep 22, 2026Contributor·

  ·
  1