[WIP] standalone shared library for custom op using ORTX-extensions infrastructure#433

Draft

jslhcl wants to merge 17 commits intomainfrom leca/customop

+1,253-2

Commits on May 10, 2024

Commits on May 13, 2024

register custom op library
jslhcl
committed

Commits on May 14, 2024

introduce paged attention
jslhcl
committed

Commits on May 16, 2024

make custom ops inside genai, not in a standalone shared lib
jslhcl
committed

Commits on May 23, 2024

sync latest code from leca/gqa branch in ort-extension which can get some result from flash attention for prmopt mode
jslhcl
committed
merged latest main
jslhcl
committed

Commits on Jun 10, 2024

Merge branch 'main' of https://github.com/microsoft/onnxruntime-genai into leca/customop
jslhcl
committed

Commits on Jun 11, 2024

set up OCOS_USE_FLASH_ATTENTION cuda_arch and OCOS_USE_CUDA
jslhcl
committed

Commits on Jun 13, 2024

test framework for custom ops is done!
jslhcl
committed

Commits on Jun 19, 2024

new test case for decoding only
jslhcl
committed

Commits on Jun 25, 2024

sync onnxruntime-ORTX with UT working
jslhcl
committed

Commits on Jun 26, 2024

fix block_tables and context_lens to avoid unknown random error
jslhcl
committed

Commits on Jun 28, 2024

Commits on Jun 30, 2024

refactor paged attention custom op
jslhcl
committed