Commits microsoft/onnxruntime

Preserve CPU MoE input when output reuses its buffer

Copilot committed 45 minutes ago

Verified ef0ddf86

Add CPU MoE intermediate-buffer reuse regression tests

Copilot committed 1 hour ago

Verified 9e4add39

Initial plan

Copilot committed 1 hour ago

Verified 6dd6e705

[CUDA] Document INT2 numerical limits and harden QMoE tests (#33858)

tianleiwu committed 5 hours ago

Verified db4a8b1e

Bump joi from 17.13.7 to 17.13.9 in /js/react_native/e2e (#33797)

dependabot[bot] committed 6 hours ago

Verified 7f16fabd

[WebGPU] Unroll bounded NVIDIA FlashAttention reductions in WGSL (#33395)

gyagp committed 14 hours ago

Verified 1082ab76

[CUDA] Add INT2 QMoE CPU/GPU offload support and optimize spin-pool wakeup (#33787)

tianleiwu committed 19 hours ago

Verified 3eb27a03

[CUDA] Fall back to FP32 in DSA INT8 TC kernel when rescale can flush a large-value-scale accumulator (#33847)

tianleiwu committed 19 hours ago

Verified 1aa82412

[CUDA] Address review follow-ups from #33731 for QMoE offload and RouterGateUpProj (#33850)

tianleiwu committed 20 hours ago

Verified f6acc099

Build WebGPU plugin EP separately for Windows documentation generation (#33829)

edgchen1 committed 22 hours ago

Verified df4b8ac0

[CUDA] Enable Tensor-Core INT3/INT4 QMoE offload decode/prefill and 8K single-tile streaming for Qwen3.8 RC5 (#33731)

tianleiwu committed 22 hours ago

Verified ec680674

Remove unused light-server from Web E2E dependencies (#33835)

eserscor committed 22 hours ago

Verified c9c4e52b

Optimize CUDA float Pow with a guarded reciprocal square root fast path (#33662)

justinchuby committed 23 hours ago

Verified a043a5fd

[CUDA] Optimize Qwen3.8-Flash-Next DSA, Indexer, HC, and QMoE kernels for SM12x and 256K context (#33802)

tianleiwu committed 1 day ago

Verified f7e70ad5

Isolate volatile build provenance from shared configuration (#33687)

justinchuby committed 1 day ago

Verified b2021624

Fix cmd escaping in Windows CMake install validation (#33840)

eserscor committed 1 day ago

Verified 07354eaf

Add vcpkg binary caching to Azure Pipelines (#33691)

eserscor committed 1 day ago

Verified 643d6ba7

[CUDA] Add BitwiseAnd, BitwiseOr, and BitwiseNot integer kernels (#33646)

justinchuby committed 1 day ago

Verified 242b1a91

Make misspell lint actionable: PR-only, blocking, no false positives (#33779)

adrastogi committed 1 day ago

Verified df5ef89b

Fix pooled GQA model buffer detachment in WebAssembly proxy tests (#33838)

eserscor committed 1 day ago

Verified f93e05cf

Handle kernel type string resolution errors without throwing (#33828)

mustjab committed 1 day ago

Verified 1aae736a

Enable macOS compiler caching with shared sccache actions (#33811)

eserscor committed 1 day ago

Verified 51aba5e4

[CUDA] Add 8-route grouped INT3 offload prefill, zero-cold doorbell short-circuit, and S=8 streaming optimizations (#33629)

tianleiwu committed 1 day ago

Verified 5049049d

Validate LinearAttention head dimensions (#33821)

apsonawane committed 1 day ago

Verified 19bcbb97

Add EP device memory capacity query API with CUDA support (#33447)

selenayang888 committed 1 day ago

Verified 5f141934

[WebGPU] Fuse rank-32 quantized LoRA with MatMulNBits (#33370)

qjia7 committed 1 day ago

Verified e64a8ae8

Fix spelling typos flagged by misspell linter (#33780)

adrastogi committed 1 day ago

Verified 22c72f53

[CPU] Support right-padded prefill chunks in GroupQueryAttention (#33505)

tianleiwu committed 1 day ago

Verified 3fc465dc

Use local Detox CLI in React Native CI (#33813)

eserscor committed 1 day ago

Verified ee239433

Validate Attention head counts (#33498)

apsonawane committed 1 day ago

Verified 7059e239