Generate: detect special architectures when loaded from PEFT #24198
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
4 tasks
amyeroberts
approved these changes
Jun 12, 2023
src/transformers/generation/utils.py
Outdated
| @@ -4512,7 +4517,8 @@ def _crop_past_key_values(model, past_key_values, maximum_length): | |||
| ) | |||
| ) | |||
| past_key_values = tuple(new_past) | |||
| elif "bloom" in model.__class__.__name__.lower(): # bloom is special | |||
| # bloom is special | |||
| elif "bloom" in model.__class__.__name__.lower() or "bloom" in model.config.architectures[0].lower(): | |||
There was a problem hiding this comment.
Are we guaranteed it's always the 0th indexed architecture?
There was a problem hiding this comment.
I have never seen more than 1 item in architectures -- in fact, the only PT reference writing into .config.architrectures is this one
I am assuming it works in the vast majority of cases -- even if the user decides to append more info into this config member, it should work.
Not a super satisfactory answer, I know 😅
src/transformers/generation/utils.py
Outdated
| @@ -4521,7 +4527,8 @@ def _crop_past_key_values(model, past_key_values, maximum_length): | |||
| ) | |||
| ) | |||
| past_key_values = tuple(new_past) | |||
| elif "gptbigcode" in model.__class__.__name__.lower(): # gptbigcode is too | |||
| # gptbigcode is too | |||
| elif "gptbigcode" in model.__class__.__name__.lower() or "gptbigcode" in model.config.architectures[0].lower(): | |||
There was a problem hiding this comment.
Same here re 0 indexing
|
The documentation is not available anymore as the PR was closed or merged. |
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. |
novice03
pushed a commit
to novice03/transformers
that referenced
this pull request
Jun 23, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Fixes #23686
As identified in this comment, a PEFT-loaded BLOOM can't be used as an assistant with assisted generation.
BLOOM (and GPTBigCode) need special handling due to their different cache API, and the architecture detection code was incompatible with PEFT models. This PR adds the logic to detect these special architectures when loaded with PEFT.