Commits microsoft/onnxruntime

[CUDA][MoE] Prioritize ready experts in fused cold-expert dispatch

BBaditi committed 1 day ago

cd12cd9f

[CUDA] Optimize INT4 QMoE offload prefill/decode and S=8 doorbell execution

tianleiwu committed 1 day ago

Verified c4b9b5a9

[CUDA] Add INT4 and INT3 QMoE CPU/GPU offload support

tianleiwu committed 1 day ago

Verified e6226966

[CUDA] Add portable QMoEOffload backend with doorbell decode, adaptive promotion, and DMA prefill

tianleiwu committed 1 day ago

Verified 635c85ab

[CUDA] Add QMoE offload kernels, spin pool, callback timing, and QMoE v4 admission hook

tianleiwu committed 1 day ago

Verified 0ac4055d

[CUDA] Add QMoESessionController and CUDA EP / Plugin EP run & graph-capture lifecycle hooks

tianleiwu committed 1 day ago

Verified 9cd70f33

[CUDA] Add session-scoped MoE/QMoE offload planner and device/pinned memory budget ledger

tianleiwu committed 1 day ago

Verified ad9add5a

fix(test,docs): skip unavailable QUICK_BUILD CUDA BF16 QMoE runner and update OperatorKernels.md

tianleiwu committed 1 day ago

Verified 5e14bce6

fix(mlas): restore QMoE FP16 rounding helpers

tianleiwu committed 1 day ago

Verified e45bd154

[CPU/MLAS] Add QMoE shared layout NVFP4/MXFP4 AVX-512 kernels and shared layout tests

tianleiwu committed 1 day ago

Verified d86249ab

[CPU/MLAS] Add QMoE shared layout INT3 AVX-512/AVX2 kernels and direct CPU QMoE

tianleiwu committed 1 day ago

Verified cf003059

[CPU/MLAS] Add QMoE shared layout INT4 AVX-512/AVX2 kernels, FileExpertStore, and direct CPU QMoE

tianleiwu committed 1 day ago

Verified 05a22ab8

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

xhcao committed 1 day ago

Verified 9735bff3

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

tianleiwu committed 1 day ago

Verified f516359e

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

Copilot committed 1 day ago

Verified 74bda515

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

tianleiwu committed 2 days ago

Verified 825a0947

[CUDA] QMoE: select the 64-row CTA tile for repacked INT3 prefill when experts average >= 24 rows (#33714)

tianleiwu committed 2 days ago

Verified 68dda2d9

Simplify 1DS shutdown ordering around the Env singleton (#33321)

bmehta001 committed 2 days ago

Verified da0422cb

Validate runtime tensor ranks in CPU Conv, ConvInteger, QLinearConv, CropAndResize, Einsum, and AffineGrid kernels (#33517)

professor-moody committed 2 days ago

Verified 9265e05a

Reject overflowing Concat axis dimensions (#33235)

xadupre committed 2 days ago

Verified f809e08b

Cpu a16 nbits large block (#32885)

mirounga committed 2 days ago

Verified 1f332641

Fix CPU Range integer boundary counts and final increment overflow (#33399)

cboseak committed 2 days ago

Verified 7fa20e3c

[CPU] Prevent QLinearSoftmax fusion with non-positive input scale (#33568)

vinay-852 committed 2 days ago

Verified 6eb43036

Fix Gemm zero-K bias scaling and signed zero (#33669)

iBreaker committed 2 days ago

Verified 9d5d2f45

Add composable C++ graph pattern rewriter (#33631)

gramalingam committed 2 days ago

Verified 9d3dc812

Bump qs and body-parser in /js/web (#33661)

dependabot[bot] committed 2 days ago

Verified 5463bcc3

Fix nightly WebGPU shader key validation check (#33693)

edgchen1 committed 2 days ago

Verified 1b66214f

Bump actions/setup-node from 6.5.0 to 7.1.0 (#33711)

dependabot[bot] committed 2 days ago

Verified a371cbde

Bump the codeql group with 4 updates (#33708)

dependabot[bot] committed 2 days ago

Verified 13762af8

Bump reviewdog/action-shellcheck from 1.32.1 to 1.34.1 (#33709)

dependabot[bot] committed 2 days ago

Verified 6485f09c