microsoft/onnxruntime

Pull Requests Commits

Support buffer sharing for DeepSeek V4 attention

Copilot committed 80 days ago

510de8b2

Refactor DeepSeek V4 contrib operators

Copilot committed 81 days ago

a314e29e

Add CPU kernels for DeepSeek V4 mix ops

Copilot committed 82 days ago

62b4e16a

Test DeepSeek V4 mix and Sinkhorn ops

Copilot committed 82 days ago

88e356d2

Add HyperConnectionMix op

tianleiwu committed 83 days ago

d45698b6

Add com.microsoft.SinkhornNormalize CUDA op

tianleiwu committed 83 days ago

14dd43d6

Add HyperConnection and HyperHead contrib ops

Copilot committed 83 days ago

2a0bb172

Split DeepSeek V4 attention into contrib ops

Copilot committed 83 days ago

8d379964

Add CUDA support for DeepSeek V4 compressed attention

Copilot committed 83 days ago

293e3b47

Add DeepSeek v4 CPU compression coverage

Copilot committed 83 days ago

Verified c15f5931

Implement CPU DeepSeek v4 compression states

Copilot committed 83 days ago

Verified c35eb4ce

Handle router_weights with CUDA sparse-mixer MoE path

Copilot committed 83 days ago

Verified 65cf7830

Fix: add narrow.h include to CUDA deepseek_v4_attention.cc

Copilot committed 84 days ago

Verified 01dc50e8

Implement DeepSeekV4Attention CPU math path

Copilot committed 84 days ago

Verified a2f459b0

Add DeepSeekV4Attention op schema and kernel scaffolding

Copilot committed 84 days ago

Verified 2472d7d5

Add DeepSeek V4 router_weights support for MoE and QMoE (CPU + CUDA)

Copilot committed 84 days ago

Verified 714bb298

Add DeepSeek V4 trailing RoPE + output de-rotation to GroupQueryAttention

Copilot committed 84 days ago

Verified 6706ac71

[CUDA] Block-scaled MatMul decode GEMV: tensor cores, packed FP4 decode and M-tiling (#31155)

tianleiwu committed 86 days ago

Verified 86af0f16

Fix LinearAttention output shape inference for standard GQA (q > kv) (#29892)

mirounga committed 86 days ago

Verified b725f3d3

Correct the overstated exception-free claim in ReadEpContextData (#31195)

GopalakrishnanN committed 86 days ago

Verified 7d3b4d93

[OpenVINO EP] Support dot-separated KV cache tensor names in stateful (CausalLM) transform (#29895)

zlma7001 committed 86 days ago

Verified 2aa23ccd

Add AzureContainerRegistry network isolation policy to WebGPU/CUDA plugin EP package test pipelines (#31194)

edgchen1 committed 86 days ago

Verified b4d126e4

[XNNPACK EP] Read Gemm M from the input tensor at Compute time (#31189)

guptaishaan committed 86 days ago

Verified 4c4d4923

[WebGPU EP] Support device-free compile-only sessions for offline graph transformation (#29681)

mingmingtasd committed 87 days ago

Verified 56071e12

[WebNN EP] Fix MatMulNBits when K is not a multiple of block_size (#31152)

Honry committed 87 days ago

Verified 38dee6cf

[CUDA] QMoE: release raw MXFP4 weight initializers after PrePack (#31154)

tianleiwu committed 87 days ago

Verified a9851c76

[CUDA] Cut the MoE routing prologue/epilogue cost for small-batch decode (#31156)

tianleiwu committed 87 days ago

Verified 69d74307

EPContext sample helper follow-ups + compiled-model encryption design doc (#29294)

GopalakrishnanN committed 87 days ago

Verified 635c7750

[QNN-EP] Add reshape handler for split axes reshape (#29660)

quic-muchhsu committed 87 days ago

Verified 3a75382d

upgrades adm-zip for onnxruntime-node (#31192)

prathikr committed 87 days ago

Verified 00e575d5

Older