microsoft/onnxruntime

Pull Requests Commits

Copilot committed 81 days ago

Verified 721fd39d

Fix #31573, prevent ARM64 SymmQgemm int16 overflow (#32057)

sylvesterkaczmarek committed 81 days ago

Verified f73b9ef8

[Build] Update cuda plugin linux aarch64 parallel to 8 (#32165)

tianleiwu committed 81 days ago

Verified 00b6cdb5

Use commit timestamps for plugin EP dev versions (#32095)

edgchen1 committed 81 days ago

Verified 84e87b4b

[CUDA] upgrade cutlass 4.7 and cudnn-frontend 1.27 (#32111)

tianleiwu committed 81 days ago

Verified fccfa10c

[WebGPU] Enable GeluFusion and BiasGeluFusion for the WebGPU EP (#32053)

fanchenkong1 committed 81 days ago

Verified 10ea381c

[Build] FP4 QMoE in Windows (#32163)

tianleiwu committed 81 days ago

Verified f1d07f0d

Add CPU and WebGPU kernels for Qwen-3.5 ops (#32106)

Copilot committed 82 days ago

Verified 0c32e39a

Bump cuda plugin ep version to 0.2 (#31970)

tianleiwu committed 82 days ago

Verified f37adf82

[CUDA] Vectorize the NVFP4 weight dequantization for prefill (#32128)

tianleiwu committed 82 days ago

Verified 2d211f8c

[CUDA] Bound the FP8 weight dequant scratch by tiling over N (#32129)

tianleiwu committed 82 days ago

Verified 8b9258f1

[WebGPU] Zero-initialize writable device allocator buffers (#32063)

qjia7 committed 82 days ago

Verified 1e9aee07

Enable split-KV for paged FlashAttention decode (#32102)

baijumeswani committed 82 days ago

Verified 0d291bc5

Improve MLAS NCHWc conv thread utilization via cost-weighted work par… (#31660)

mirounga committed 83 days ago

Verified c2298d30

Use native block tables for 128-token XQA pages (#32127)

baijumeswani committed 83 days ago

Verified fab46917

Implement WebGPU subgroup-size-control infrastructure (#32056)

jchen10 committed 83 days ago

Verified fdd011e0

Fix/ort runtime optimization trust (#32077)

apsonawane committed 84 days ago

Verified dd64c8ad

Fix/pool negative pad validation (#32076)

apsonawane committed 84 days ago

Verified aa73c1e7

Add group size 6 support to paged XQA (#32108)

baijumeswani committed 85 days ago

Verified e858164c

[CUDA] Parallelize ArgMax/ArgMin over wide last axes (#32092)

justinchuby committed 85 days ago

Verified d4d792f5

Improve PagedAttention dispatch diagnostics (#32099)

baijumeswani committed 85 days ago

Verified a2c0007a

[CUDA] Enable FP4 QMoE by default (#32096)

tianleiwu committed 85 days ago

Verified aac951db

Bound QMoE workspace with configurable row tiling (#32097)

baijumeswani committed 86 days ago

Verified c5300b71

Support bidirectional GroupQueryAttention on CPU and CUDA (#31704)

Copilot committed 86 days ago

Verified 125ea215

Enable WebGPU CI for WebGPU plugin EP release branches (#32090)

edgchen1 committed 86 days ago

Verified 48163990

Validate generation subgraph shapes (#32078)

apsonawane committed 86 days ago

Verified 337ec523

Validate CUDA GatherElements count (#32030)

apsonawane committed 86 days ago

Verified 2b9a9480

Validate in-memory initializer references (#32042)

apsonawane committed 86 days ago

Verified 85d92926

Fix/qdq optional zero point input (#32051)

apsonawane committed 86 days ago

Verified 927b53ae

Validate MatMulFpQ4 shape inputs (#32032)

apsonawane committed 86 days ago

Verified 0da10ef2

Older