Pull Requests microsoft/onnxruntime

[CUDA] MatMulBlockQuantizedFp8Weight: fold the W8A8 activation QDQ into the decode GEMV

#31481 opened 2026-08-02 07:17 by tianleiwu

[CUDA] GroupQueryAttention: dequantize the K and V caches in one launch

#31480 opened 2026-08-02 07:14 by tianleiwu

[CUDA] Skip FP4 QMoE fc1 activation expansion

#31479 opened 2026-08-02 02:31 by tianleiwu

[CUDA] MatMul: add an opt-in split-K GEMV for small-N fp16 shapes

#31478 opened 2026-08-01 19:54 by tianleiwu

[CUDA] Fix Abs signed zero handling

#31477 opened 2026-08-01 14:04 by MohamedElashri

[CUDA] Speed up 8-bit MatMulNBits dequantization with byte permutes

#31350 opened 2026-08-01 00:52 by tianleiwu

[CUDA] Coalesce QMoE MXFP4/NVFP4 weight dequantization

#31349 opened 2026-07-31 23:55 by tianleiwu

[Compile API] Plan for adding support for writing external initializers to buffer

#31347 opened 2026-07-31 22:22 by skottmckay

Ensure ONNX and Protobuf use compatible package configurations

#31346 opened 2026-07-31 22:15 by Copilot

Add AzureContainerRegistry to network isolation policy for additional pipelines.

#31344 opened 2026-07-31 17:36 by edgchen1

[Shape Inference] Add shape inference for MatMulIntegerToFloat op

#31199 opened 2026-07-31 15:58 by roberto-laudani

[WebNN EP] Support LpNormalization op

#31197 opened 2026-07-31 06:55 by Honry

Make the tellg failure check explicit and document the EPContext write gate error handling

#31196 opened 2026-07-31 03:04 by GopalakrishnanN

[CUDA] Port ShardedMoE to the new MoE GEMM backend

#31191 opened 2026-07-30 19:09 by tianleiwu

Fix shape inference failure with file-backed external data

#31169 opened 2026-07-30 13:00 by yoonseok-kim

[CUDA] Speed up the NVFP4 QMoE decode GEMV and enable it for MTP verify

#31159 opened 2026-07-30 09:36 by tianleiwu

[CUDA] Add LinearAttentionGate and GatedRMSNorm contrib operators

#31158 opened 2026-07-30 09:36 by tianleiwu

[CUDA] Add state_window to LinearAttention and CausalConvWithState to support MTP

#31157 opened 2026-07-30 09:36 by tianleiwu

[WebNN EP] Reuse shared WASM loader for Blob-backed external data

#31151 opened 2026-07-30 05:56 by Honry

Skip redundant present_key/value copy when aliased to external KV cache (follow-up to #29715) ep:CUDA

#31150 opened 2026-07-30 04:57 by titaiwangms

Throw error on negative Split tensor axis [CPU]

#31149 opened 2026-07-30 00:03 by nenad1002

Fix LayerNorm fusion selecting the Div output as scale or bias

#31148 opened 2026-07-29 23:57 by Sammy-Dabbas

[ARM] MLAS: SVE i8mm (svmmla) int8 QGEMM kernels, portable machine code

#31146 opened 2026-07-29 21:54 by mirounga

[ARM] MLAS: portable machine-code SVE elementwise kernels, FEXPA exp

#31145 opened 2026-07-29 21:53 by mirounga

Do not fuse LayerNorm when ReduceMean has keepdims=0

#31144 opened 2026-07-29 20:01 by guptaishaan

[ARM] MLAS: route fp32 SGEMM to the KleidiAI SVE ukernel on SVE-only

#31143 opened 2026-07-29 18:10 by mirounga

Fix segfault in MatMulNBits prepacking when weights/scales are parent-graph initializers in subgraph

#31141 opened 2026-07-29 15:51 by Copilot

[OpenVINO EP] Fix get/set_float_initializer_data for raw_data backed float initializers

#31138 opened 2026-07-29 11:46 by wangw-1991

Auto-generated baselines by 1ES Pipeline Templates

#31136 opened 2026-07-29 11:03 by microsoft-github-policy-service[bot]

[WebGPU EP] Support int64 for Tile and Concat ep:WebGPU

#31049 opened 2026-07-29 08:50 by miaobin