PEFT: Access active_adapters as a property in Trainer

[pashminacameron]

PEFTModel has a property called active_adapters. It was being accessed as a function which led to the below error.
Seeing issues such as

    trainer.train()
  File "C:\dev\anaconda3\envs\env-gpu\Lib\site-packages\transformers\trainer.py", line 1885, in train
    return inner_training_loop(
           ^^^^^^^^^^^^^^^^^^^^
  File "C:\dev\anaconda3\envs\env-gpu\Lib\site-packages\transformers\trainer.py", line 2339, in _inner_training_loop
    self._load_best_model()
  File "C:\dev\anaconda3\envs\env-gpu\Lib\site-packages\transformers\trainer.py", line 2629, in _load_best_model
    active_adapter = model.active_adapters()[0]
                     ^^^^^^^^^^^^^^^^^^^^^^^
TypeError: 'list' object is not callable

Issue seems similar to #30754 and was likely introduced in #30738

Fixes: #30754

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
@younesbelkada

[younesbelkada]

Thanks a lot for fixing !

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[amyeroberts]

Thanks for the quick fix!

[ghost]

Thanks :) now I can stop hitting my head on the table.

[speculaas]

I thought this was the solution to my similar error,
but it turned out, my transformer actually contains this commit but failed in a different way:

File "/home/jimmy/pyvenv/310/lib/python3.10/site-packages/transformers/trainer.py", line 2694, in _load_best_model
active_adapter = model.active_adapters[0]
TypeError: 'method' object is not subscriptable

I am trying to run this example:
https://github.com/huggingface/peft/blob/fb607d00adc7b8aa1694f58a786ce2cd5681710d/docs/source/task_guides/ptuning-seq-classification.md

Thanks huggingface for these libraries!

[amyeroberts]

cc @BenjaminBossan

[pashminacameron]

Hey @speculaas Which peft and transformers versions (or commits) do you have? Thanks.

[speculaas]

thanks for your help!

My peft and transformers versions are:

.venv/lib/python3.12/site-packages/peft$ pgrep "_version" 
./__init__.py:20:__version__ = "0.11.1"

.venv/lib/python3.12/site-packages/transformers$ pgrep "_version" | grep _init 
./__init__.py:21:__version__ = "4.42.3"

[BenjaminBossan]

Thanks for reporting the problem. The problematic line is the following:

https://github.com/huggingface/peft/blob/1b16753a6aca2fc9581823c8e2efd628b7d2b676/src/peft/peft_model.py#L181

The issue stems from the fact that if the base_model is some kind of PEFT model, e.g. LoraModel, self.base_model.active_adapters will return a list. However, if the underlying base model is a transformers model, as is the case here, active_adapters is a method that needs to be called. It is unfortunate that there are two methods with the same name but one is a property and the other is not.

Anyway, this is an issue that should be fixed on the PEFT side. I'll work on a PR.

[BenjaminBossan]

PR: huggingface/peft#1975

[BenjaminBossan]

The PR is merged.

[speculaas]

Thanks for the fix and explanation!

[speculaas]

The PR is merged.

Do you recommend install it from source?

[BenjaminBossan]

Do you recommend install it from source?

Installing from source is always a little bit more risky when it comes to breaking things, so if you can wait, I would wait for the next release. That said, if you fix the commit hash, you'll be most likely fine when installing from source.