cloudhan 51b67fa15c Make ROCm Attention biased+masked and biased+nomask scaling logic consistent (#14976) ... The biased+masked and biased+nomask have different scaling logic in current ROCm implementation Currently, biased + masked: (QK'+ bias) * scale + convert(mask) biased + nomask: QK' * scale + bias which is not correct. What we want is QK' * scale [+ bias] That is, bias should not be scaled. This effectively follows https://github.com/microsoft/onnxruntime/pull/14517/files?w=1#diff-e4768ce15a73499f584f9cd7d71adcb1ff2ed8d68ad7e496723a4775cbc35e33		2023-03-09 23:37:50 +08:00
..
contrib_ops	Make ROCm Attention biased+masked and biased+nomask scaling logic consistent (#14976)	2023-03-09 23:37:50 +08:00
core	fix rocBLAS extensions API issue; add batched- and strided_batched- cases (#14883)	2023-03-09 23:23:35 +08:00
python	[ROCm] SkipLayerNorm: add more configs for block size; loosen constraints (#14900)	2023-03-09 22:27:01 +08:00
test	[CUDA] Support decoding multihead self-attention implementation (#14848)	2023-03-08 09:17:54 -08:00
tool/etw
wasm	[js/web] support flag 'optimizedModelFilePath' in session options (#14355)	2023-02-24 15:50:15 -08:00
__init__.py	Add GetVersionSting API for C++, C# and Python (#14873)	2023-03-02 17:11:07 -08:00
ReformatSource.ps1
ReformatSourcePython.bat
VSCodeCoverage.runsettings