microsoft/onnxruntime

Pull Requests Commits

Merge main integration through asonawane/ported-ops-sparse-attention-indexer

apsonawane committed 26 days ago

abf747b4

Merge main integration through asonawane/ported-ops-gdn-convolution

apsonawane committed 26 days ago

222dfecc

Merge main integration through asonawane/ported-ops-hyperconnection-tensor

apsonawane committed 26 days ago

9d0678fb

Merge main integration through asonawane/ported-ops-normalization-rotary

apsonawane committed 26 days ago

8a788aa8

Merge main into normalization and rotary stack

apsonawane committed 26 days ago

26cb2368

Merge updated prerequisite asonawane/ported-ops-sparse-attention-indexer

apsonawane committed 26 days ago

854835bc

Merge updated prerequisite asonawane/ported-ops-gdn-convolution

apsonawane committed 26 days ago

35a4d3bc

Merge updated prerequisite asonawane/ported-ops-hyperconnection-tensor

apsonawane committed 26 days ago

78820657

Merge updated prerequisite asonawane/ported-ops-normalization-rotary

apsonawane committed 26 days ago

6e0540fb

apsonawane committed 26 days ago

7347338e

apsonawane committed 26 days ago

036b1db4

apsonawane committed 26 days ago

c64cce88

apsonawane committed 26 days ago

54870655

apsonawane committed 26 days ago

d76d17c0

Allow onnxruntime_DISABLE_EXCEPTIONS without onnxruntime_MINIMAL_BUILD (#33120)

rvandermeulen committed 26 days ago

Verified 9a662b93

Optimize quantized MoE and block-scaled FP8 CUDA projections

apsonawane committed 26 days ago

9e8a15a5

Extend sparse attention caches and optimize packed indexing

apsonawane committed 26 days ago

9b8aaa0c

Extend GDN and causal convolution fused state handling

apsonawane committed 26 days ago

62dbc36f

Add fused hyper-connection and tensor CUDA operators

apsonawane committed 26 days ago

437320d9

Add exact normalization and rotary CUDA operator modes

apsonawane committed 26 days ago

77642733

[CUDA] Add opt-in tail-safe INT4 MatMulNBits dispatch for N % 8 == 4, M <= 4 (#33299)

tianleiwu committed 26 days ago

Verified bb68f5de

Skip MatMul BatchNormalization fusion when the channel axis moves (#33392)

samridh10exe committed 27 days ago

Verified dc1ae7a1

[CUDA] Add decode GEMV kernel for K-packed NVFP4 QMoE (#33354)

tianleiwu committed 27 days ago

Verified 69eeb2bb

[CUDA] Add opt-in native E4M3 block-scale conversion to NVFP4 GEMV (#33323)

sayanshaw24 committed 27 days ago

Verified 9eec7d5f

Fix WebGPU Conv2dMM input indexing for vec3 channels (#33432)

gyagp committed 27 days ago

Verified 0614f48d

[WebGPU] Support block-quantized FP8 dense MatMul (#33074)

sayanshaw24 committed 27 days ago

Verified 81f5efab

Revert "Fix non-deterministic constant folding output" (#32471)

xiaofeihan1 committed 27 days ago

Verified c43d712e

Fix safety guards in training loss fusions (#32987)

Yuhx141 committed 28 days ago

Verified c1ea5bff

[CUDA plugin EP] Improve NuGet package metadata (#33405)

tianleiwu committed 28 days ago

Verified ce1d73c2

Fix 1DS telemetry deadlock during Windows process termination (#33291)

qjia7 committed 29 days ago

Verified c64c9fd7

Older