Repository navigation

Sync with Microsoft ONNX Runtime - 02112026 - #1379

Open

ai-fw-intg wants to merge 31 commits into

ovep-developfrom

sync_msft_02112026

Open

Sync with Microsoft ONNX Runtime - 02112026#1379
ai-fw-intg wants to merge 31 commits into
ovep-developfrom
sync_msft_02112026

Commits on Oct 29, 2026
- [CUDA] HyperConnectionLowRankMix: fused prefill PreMix+Inject+PostGate and 4-lane warp-shuffle decode epilogue (#33738)
  
  tianleiwu
  and
  Copilot
  authored
  825a094
- Design EP-registered runtime fusion through capability recipes (#33404)
  
  authored
  74bda51
Commits on Oct 30, 2026
- [CUDA] Add SwiGLU contrib operator and SwiGLU fusion (#32126)
  
  tianleiwu
  and
  Copilot
  authored
  f516359
- webgpu: using subgroup matrix to optimize Conv operator (#32252)
  xhcao
  authored
  9735bff
- Bump onnx from 1.21.0 to 1.22.0 in /onnxruntime/python/tools/transformers/models/stable_diffusion/requirements/cuda12 (#33734)
  dependabot[bot]
  authored
  b5fe413
- Skip QDQ node group rewrites for float8/float4 quantized types (#33663)
  
  sisleyli
  and
  cursoragent
  authored
  3f975f9
- Exclude filesystem-dependent pattern rewriter tests from WASM (#33745)
  
  edgchen1
  and
  Copilot
  authored
  8b8e5dd
- [WebGPU] Add QuantizeLinear and int8 Cast (#33683)
  
  adrastogi
  and
  Copilot
  authored
  1a440cf
- [WebGPU] Add LpPool and channels-last bilinear Resize (#33679)
  
  adrastogi
  and
  Copilot
  authored
  529206c
- [Pilot] Add region-scoped pattern rewriting through EP capabilities (#33652)
  
  authored
  14663eb
- Keep a system cuDNN include dir out of the CUDA plugin's include list (#33736)
  arnej27959
  authored
  4605d0b
- Add CUDA BitShift integer kernels (#33647)
  
  justinchuby
  and
  Copilot
  authored
  103d010
- Make WebGPU CI validation failures blocking (#33715)
  
  edgchen1
  and
  Copilot
  authored
  9016c6d
- Schedule daily CUDA 13 combined wheel builds (#33771)
  
  eserscor
  and
  Copilot
  authored
  dd543fe
- Add CACHE when unsetting CMAKE_CUDA_ARCHITECTURES to ensure it's empty (#28019)
  jmcarcell
  authored
  c719d1f
- [WebGPU] Add elementwise kernels and integer types for WebNN conformance (#33680)
  
  authored
  c46c2a1
- Add CUDA BitwiseXor integer kernels (#33645)
  
  justinchuby
  and
  Copilot
  authored
  94f6132
- [CUDA] Add sm_121 to XQA arch-specific branches (#33762)
  
  tianleiwu
  and
  Copilot
  authored
  13c2908
- Fix XQA reduction races in shared-cache GQA decoding (#33703)
  
  authored
  e7cc17f
- [CPU/MLAS] Add QMoE shared layout INT4/INT3/FP4 AVX-512 & AVX2 kernels, FileExpertStore, and direct CPU QMoE (#33450)
  
  tianleiwu
  and
  Copilot
  authored
  c574f2d
- [CUDA] Add RouterGateUpProj contrib operator (#33650)
  
  tianleiwu
  and
  Copilot
  authored
  f5ff8ca
Commits on Oct 31, 2026
Commits on Nov 1, 2026

Navigation Menu

Repository navigation

Sync with Microsoft ONNX Runtime - 02112026 - #1379

Sync with Microsoft ONNX Runtime - 02112026#1379
ai-fw-intg wants to merge 31 commits into
ovep-developfrom
sync_msft_02112026

Commits on Oct 29, 2026

[CUDA] HyperConnectionLowRankMix: fused prefill PreMix+Inject+PostGate and 4-lane warp-shuffle decode epilogue (#33738)

Design EP-registered runtime fusion through capability recipes (#33404)

Commits on Oct 30, 2026

[CUDA] Add SwiGLU contrib operator and SwiGLU fusion (#32126)

webgpu: using subgroup matrix to optimize Conv operator (#32252)

Bump onnx from 1.21.0 to 1.22.0 in /onnxruntime/python/tools/transformers/models/stable_diffusion/requirements/cuda12 (#33734)

Skip QDQ node group rewrites for float8/float4 quantized types (#33663)

Exclude filesystem-dependent pattern rewriter tests from WASM (#33745)

[WebGPU] Add QuantizeLinear and int8 Cast (#33683)

[WebGPU] Add LpPool and channels-last bilinear Resize (#33679)

[Pilot] Add region-scoped pattern rewriting through EP capabilities (#33652)

Keep a system cuDNN include dir out of the CUDA plugin's include list (#33736)

Add CUDA BitShift integer kernels (#33647)

Make WebGPU CI validation failures blocking (#33715)

Schedule daily CUDA 13 combined wheel builds (#33771)

Add CACHE when unsetting CMAKE_CUDA_ARCHITECTURES to ensure it's empty (#28019)

[WebGPU] Add elementwise kernels and integer types for WebNN conformance (#33680)

Add CUDA BitwiseXor integer kernels (#33645)

[CUDA] Add sm_121 to XQA arch-specific branches (#33762)

Fix XQA reduction races in shared-cache GQA decoding (#33703)

[CPU/MLAS] Add QMoE shared layout INT4/INT3/FP4 AVX-512 & AVX2 kernels, FileExpertStore, and direct CPU QMoE (#33450)

[CUDA] Add RouterGateUpProj contrib operator (#33650)

Commits on Oct 31, 2026

[WebGPU] Use segment accessors in SkipLayerNormalization (#33760)

[CUDA] Add session-scoped MoE/QMoE offload planner, lifecycle controller, and memory budget ledger (#33451)

[CUDA] Add H256 BF16-query INT8-cache group-6/16 paged speculative XQA (#33538)

[CUDA] Pass cached SM version to cuDNN frontend Conv graphs (#33770)

[WebGPU] Support float16 GRU and LSTM (#33678)

[CUDA] Tensor-core QMoE INT4/INT3 decode, INT8 MatMulNBits GEMV and large-M dequantize + cuBLAS GEMM (#33689)

Commits on Nov 1, 2026

[WebGPU] Evaluate Erf, Exp, Tan, f16 reductions and InstanceNormalization in f32 (#33682)

[CUDA] Tensor-core INT8-KV DynamicSparseAttention prefill and decode, branch-free hyper-connection sigmoid (#33707)

[CUDA] Add portable QMoE CPU/GPU offload backend with doorbell decode, adaptive promotion, and DMA prefill (#33452)

Merge remote-tracking branch 'origin/master' into sync_msft_02112026