ImportError: cannot import name 'VBLoRAConfig' from 'peft'

[KQDtianxiaK]

System Info

RTX 4090D(24GB)
peft == 0.12.0
transformers == 4.44.0
Has this method been removed?

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder
• My own task or dataset (give details below)

Reproduction

from peft import VBLoRAConfig, get_peft_model, TaskType, AutoPeftModelForSequenceClassification
model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2)
config = VBLoRAConfig( task_type=TaskType.SEQ_CLS, target_modules=['key','value','query'], num_vectors=60, vector_length=256, save_only_topk_weights=Flase )
model_VBLoRA = get_peft_model(model, config)

Expected behavior

Get an answer to solve this problem

[BenjaminBossan]

VB-LoRA has been merged very recently and thus is not yet part of any PEFT release. You can either wait until the next PEFT release (my estimation is within the next 2 weeks) or you install PEFT from source: python -m pip install git+https://github.com/huggingface/peft.

[KQDtianxiaK]

VB-LoRA has been merged very recently and thus is not yet part of any PEFT release. You can either wait until the next PEFT release (my estimation is within the next 2 weeks) or you install PEFT from source: python -m pip install git+https://github.com/huggingface/peft.

Thank you for your prompt and accurate reply, I will look forward to the next version.
In addition, I have another question: why the peft library does not have adapters proposed by the paper "Parameter-efficient transfer learning for NLP" so far?

[BenjaminBossan]

why the peft library does not have adapters proposed by the paper "Parameter-efficient transfer learning for NLP" so far?

The answer to this type of question is: Because nobody created a PR to add it :) Yes, sometimes the PEFT maintainers (which is just me at the moment) will go ahead and implement a particularly promising method. But in general this takes time that we don't have, so we mostly rely on community contributions.

See also #244.

[KQDtianxiaK]

why the peft library does not have adapters proposed by the paper "Parameter-efficient transfer learning for NLP" so far?

The answer to this type of question is: Because nobody created a PR to add it :) Yes, sometimes the PEFT maintainers (which is just me at the moment) will go ahead and implement a particularly promising method. But in general this takes time that we don't have, so we mostly rely on community contributions.

See also #244.

Thanks for your answers and all the contributors, hope one day I can be one of you.