microsoft/onnxruntime

Pull Requests Commits

Add executive summary to the report

Copilot committed 68 days ago

ea50c1ee

Reframe size attribution for planned EP isolation; show EP JS glue can ship with the EP

Copilot committed 68 days ago

6ce7ca73

Address review: retract the JSPI-only conclusion, fix harness hygiene, bound the size claim

Copilot committed 68 days ago

0ca01e20

Drop the heap-ownership probe: out of scope for ORT plugin EPs

Copilot committed 68 days ago

720ea794

Make the shared-heap probe actually free across the module boundary

Copilot committed 68 days ago

7c9b764a

Prototype: onnxruntime-web loading a plugin EP as an Emscripten side module

Copilot committed 69 days ago

a25c8ac3

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

tianleiwu committed 69 days ago

Verified 2d211f8c

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

tianleiwu committed 69 days ago

Verified 8b9258f1

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

qjia7 committed 70 days ago

Verified 1e9aee07

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

baijumeswani committed 70 days ago

Verified 0d291bc5

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

mirounga committed 71 days ago

Verified c2298d30

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

baijumeswani committed 71 days ago

Verified fab46917

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

jchen10 committed 71 days ago

Verified fdd011e0

Fix/ort runtime optimization trust (#32077)

apsonawane committed 71 days ago

Verified dd64c8ad

Fix/pool negative pad validation (#32076)

apsonawane committed 71 days ago

Verified aa73c1e7

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

baijumeswani committed 72 days ago

Verified e858164c

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

justinchuby committed 73 days ago

Verified d4d792f5

Improve PagedAttention dispatch diagnostics (#32099)

baijumeswani committed 73 days ago

Verified a2c0007a

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

tianleiwu committed 73 days ago

Verified aac951db

Bound QMoE workspace with configurable row tiling (#32097)

baijumeswani committed 73 days ago

Verified c5300b71

Support bidirectional GroupQueryAttention on CPU and CUDA (#31704)

Copilot committed 73 days ago

Verified 125ea215

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

edgchen1 committed 74 days ago

Verified 48163990

Validate generation subgraph shapes (#32078)

apsonawane committed 74 days ago

Verified 337ec523

Validate CUDA GatherElements count (#32030)

apsonawane committed 74 days ago

Verified 2b9a9480

Validate in-memory initializer references (#32042)

apsonawane committed 74 days ago

Verified 85d92926

Fix/qdq optional zero point input (#32051)

apsonawane committed 74 days ago

Verified 927b53ae

Validate MatMulFpQ4 shape inputs (#32032)

apsonawane committed 74 days ago

Verified 0da10ef2

Validate CUDA QDQ element counts (#32029)

apsonawane committed 74 days ago

Verified 6a86c063

Validate GQA fusion projection shapes (#32018)

apsonawane committed 74 days ago

Verified c07c4197

Validate FastGelu fusion scale node (#32016)

apsonawane committed 74 days ago

Verified 68aa29c5

Older