-
Notifications
You must be signed in to change notification settings - Fork 66
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fix for e2e tests #927
Fix for e2e tests #927
Conversation
👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why is the session being reset in these cases and not our example cases? I'm just surprised we didn't see it then
@dsikka Here we can change to not to reset, but saving/not saving doesnt affect the test |
…oject/llm-compressor into sparseautomodel-removal-followup
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Could we update such that we're saving the recipe now please?
It's informative to have that for the e2e cases.
@dsikka |
We push the outputs and refer to them frequently. Please look at the complete test |
If we were referring to it frequently, we should have known recipe wasnt being pushed previously. |
dont merge yet please, not resetting session, so lifecycle is finializing more than once, which raises error. |
704f11c
to
a0c49e4
Compare
a0c49e4
to
5cd4c11
Compare
Follow up pr to check the contexts of the files, and to make sure recipes exist in the session is here |
SUMMARY:
Issue was caused by trying to save the recipe on modify_save_pretrained. This runs after the session is reset, so no recipe to save.
With SparseAutoModelForCausalLM before the commit, in the same test, recipe is not saved. It is saved in the save_model_and_recipe, which runs after oneshot. We dont do it here anymore because 1. if output path is not provided it doesnt save; 2. if we add again, there will be double save
[Notes]
Another way to fix this is to not to reset the session. This allows us to load and save the recipe. pre-commited SparseAutoModelForCausalLM state does not save the recipe.
TEST PLAN:
Checked that the output from oneshot is the same and pass the e2e tests.