PR #32899 Improve the app-EP weightless contract

Improve the app-EP weightless contract to allow apps to select the better fitting mode

javier-intel committed 4 days ago

Introduce new EP metadata key to report supported modes to support legacy apps, set legacy session key to support legacy EPs

javier-intel committed 4 days ago

Fix case where invalid mode was accepted when no plugin EP was used, add test external mode used on for mixed initializer models

javier-intel committed 4 days ago

Add exact weightless mode string comparison and other edge case fixes

javier-intel committed 4 days ago

Review fixes

javier-intel committed 4 days ago

Fix rebase issues, update ORT version references to 1.32

javier-intel committed 4 days ago

Remove premature 1.32 version marker and checks

javier-intel committed 4 days ago

Include missing header

javier-intel committed 4 days ago

Reject aggregate mode in legacy weightless_support key

javier-intel committed 4 days ago

Introduce context model metadata to track external dependencies

javier-intel committed 4 days ago

Add better support for case when ep.enable_weightless=1

javier-intel committed 4 days ago

Fine tune source model warning criteria

javier-intel committed 4 days ago

Track only the internal initializers count and one example name

javier-intel committed 4 days ago

Record context model dependencies for all EP API versions and include subgraph dependencies

javier-intel committed 4 days ago

Rebase fix

javier-intel committed 4 days ago

Capture dependencies for initializer-only branches and propagate plugin EP init failure

javier-intel committed 4 days ago

Adjust expected value for CudaPluginArenaTest.CudaGraphRejectsEnvironmentAllocator

javier-intel committed 4 days ago

Android minimal build size minor reductions

javier-intel committed 4 days ago

onnxruntime Improve the app-EP weightless contract #32899 Merged