Build EXTENSION_DATA_LOADER with size_test#12678
Conversation
[ghstack-poisoned]
swolchok
added
release notes: none
extension/llm/tokenizers
Outdated
| @@ -1 +1 @@ | |||
| Subproject commit ea8691c6200c702cf28dd2aa2da902e9299b6bb3 | |||
| Subproject commit f09feca15849a790c05b3b7855e7c62ce26ba94b | |||
There was a problem hiding this comment.
nope, good catch, should be in the EXPORT PR or split out
|
|
||
| check_conflicting_options_on( | ||
| IF_ON EXECUTORCH_BUILD_ARM_BAREMETAL CONFLICTS_WITH | ||
| EXECUTORCH_BUILD_EXTENSION_DATA_LOADER EXECUTORCH_BUILD_PTHREADPOOL |
There was a problem hiding this comment.
I suspect this or install_executorch used by the ci scripts was the reason it was working before?
If we want to not enable this by default then perhaps we should also update install_executorch (if not using the default preset already).
And that should also make the baremetal build fail. I also realized that we don't run the baremetal size-test binary on the CI, so we may need to fix that too.
There was a problem hiding this comment.
the reason it was working before?
you may be interested in later PRs in this stack, particularly #12744
|
build-demo-ios looks like a flake, noting CI looks good before removing the accidental tokenizers bump and starting merge of the stack |
| local EXTRA_BUILD_ARGS="${@}" | ||
|
|
||
| if [[ "$EXTRA_BUILD_ARGS" == *"-DEXECUTORCH_BUILD_ARM_BAREMETAL=ON"* ]]; then | ||
| local BUILD_DATA_LOADER="OFF" |
There was a problem hiding this comment.
Nit: is double quote needed?
There was a problem hiding this comment.
I don't think so, but I also don't think it hurts
|
merging based on previous good CI run |
size_test.cpp uses FileDataLoader. Looks like this happens to work currently because of the missing cmake_deps.toml entry that gets fixed in pytorch#12744
3 participants
size_test.cpp uses FileDataLoader. Looks like this happens to work currently because of the missing cmake_deps.toml entry that gets fixed in #12744