microsoft/onnxruntime

Pull Requests Commits

fix avx512 reduce max

yufenglee committed 1 year ago

c6b35aa4

avx511f on windows

yufenglee committed 1 year ago

426f66f7

yihonglyu committed 1 year ago

cbfa3f83

Handle unalign + align in MlasComputeSoftmaxThreaded

yihonglyu committed 1 year ago

2a1bd6c3

Function test of MlasReduceMaximumF32KernelAvx512F

yihonglyu committed 1 year ago

c317adf9

Add 64-bytes alignment prototype

yihonglyu committed 1 year ago

631e802a

Add 4, 8, 128-bytes align in bench (remove later)

yihonglyu committed 1 year ago

5e5b6e91

Revise logic of REDUCEMAXIMUMF32KERNELAVX512F

yihonglyu committed 1 year ago

821ccd19

Comment out Aligned of bench_computesoftmax.cpp

yihonglyu committed 1 year ago

cd3919dc

Support MlasMinimumF32Value

yihonglyu committed 1 year ago

ab90a9b6

Add computesoftmax bench option Aligned

yihonglyu committed 1 year ago

b635f8d6

Add softmax bench and SoftmaxKernelAvx512F.S

yihonglyu committed 1 year ago

18c1e72c

[js/webgpu] Use global id in attention and instance-norm (#20008)

axinging committed 1 year ago

Verified a2998e5d

[java][DML EP] Modifying dml_provider_factory.h so it can compile as a C header file (#20157)

Craigacp committed 1 year ago

Verified 262b6bd3

Update api backward compatibility (#20136)

xiaoyu-work committed 1 year ago

Verified 3979f53a

[CANN] Add dump_om_model flag (#20075)

wangshuai09 committed 1 year ago

Verified 3e2b659f

Fix bug related to export failure for DynamicQuantizeLSTM [issue 15465] (#20160)

dhruvbird committed 1 year ago

Verified 742d4135

optimize threading of mha (#20088)

yufenglee committed 1 year ago

Verified 91654988

Label encoder fusion (#19761)

neNasko1 committed 1 year ago

Verified 9d06e1bf

enable lto in Python-CUDA-Packaging Pipline (#20164)

mszhanyi committed 1 year ago

Verified 523ef042

[DML EP] DML Graph Serialization Bug (#19748)

sumitsays committed 1 year ago

Verified e1e292f9

Add flash attention v2 and INT4 CUDA for LLaMA E2E benchmarking (#20149)

kunal-vaishnavi committed 1 year ago

Verified a0ebd5fe

fix quantization errors of ConvTranspose with per_channel=True (#19996)

mo-ja committed 1 year ago

Verified 00244ea1

Fix MoE tensor parallelism tests (#20147)

wangyems committed 1 year ago

Verified f3a86421

Enable generic feature level devices in DML EP (#20114)

jeffbloo committed 1 year ago

Verified 2f315604

add API function GetAliasMap and ReleaseAliasMap in OrtCustomOp (#20145)

jslhcl committed 1 year ago

Verified 604b2842

fix shape inference bug (#19848)

inisis committed 1 year ago

Verified 83968458

[Quant] Fix accuracy_level config option for MatMul 4bits quantizer (#20146)

adrianlizarraga committed 1 year ago

Verified b1a5eb25

add QMoE (#20108)

wangyems committed 1 year ago

Verified 17919717

Fix transformer layer detection for recompute (#20106)

pengwa committed 1 year ago

Verified 2092bebc

Older