ChTauchmann/Prompt-Transferability

On Transferability of Prompt Tuning for Natural Language Processing

<<<<<<< Updated upstream

    # On Transferability of Prompt Tuning for Natural Language Processing

Stashed changes

Prompt Transferability

This is the source code of "On Transferability of Prompt Tuning for Natural Language Processing", an NAACL 2022 paper [pdf].

Overview

Prompt tuning (PT) is a promising parameter-efficient method to utilize extremely large pre-trained language models (PLMs), which can achieve comparable performance to full-parameter fine-tuning by only tuning a few soft prompts. However, PT requires much more training time than fine-tuning. Intuitively, knowledge transfer can help to improve the efficiency. To explore whether we can improve PT via prompt transfer, we empirically investigate the transferability of soft prompts across different downstream tasks and PLMs in this work. We find that (1) in zero-shot setting, trained soft prompts can effectively transfer to similar tasks on the same PLM and also to other PLMs with a cross-model projector trained on similar tasks; (2) when used as initialization, trained soft prompts of similar tasks and projected prompts of other PLMs can significantly accelerate training and also improve the performance of PT. Moreover, to explore what decides prompt transferability, we investigate various transferability indicators and find that the overlapping rate of activated neurons strongly reflects the transferability, which suggests how the prompts stimulate PLMs is essential. Our findings show that prompt transfer is promising for improving PT, and further research shall focus more on prompts' stimulation to PLMs.

Setups

• pip>=21.3.1
• python>=3.6.13
• torch==1.9.0+cu111

You could refer environment.yml for more details.

Requirements

bash requirement.sh

Download PLM Checkpoints from Huggieface

You could directly download PLMs (Checkpoints and Tokenizers) to the corresponding directories as below:

Bert: bert-base-uncased

mkdir BertForMaskedLM

Roberta: roberta-base, roberta-large

mkdir RobertaForMaskedLM
mkdir RobertaLargeForMaskedLM

T5: t5-small, t5-base, t5-large

mkdir T5SmallForMaskedLM
mkdir T5ForMaskedLM
mkdir T5LargeForMaskedLM

Download Downstream Dataset

Refer to data/download_dataset_convert_file.py

Train (Prompt Tuning)

Perform Prompt Tuning:

bash train.sh

Example:

gpus=0
DATASET="SST2"
BACKBONE="Roberta"

CUDA_VISIBLE_DEVICES=$gpus python3 train.py --config config/${DATASET}Prompt${BACKBONE}.config \
    --gpu $gpus \
    --checkpoint model/${DATASET}Prompt${BACKBONE} \

Main Arguments:

• --config: Prompt Tuning (PT) / Validation configurations in /config directory.
• --gpu: Assign gpu ID.
• --checkpoint: Initialize prompt parameters for Prompt Tuning (PT).

After prompt tuning, you could use copy_prompt_to_taskpromptdir.py to convert trained prompts and save them as tensors in /task_prompt_emb.

Evaluate (Trained Prompts)

Evaluate trained prompts on the corresponding tasks:

bash valid.sh

Example:

gpus=0
DATASET="SST2"
BACKBONE="Roberta"

CUDA_VISIBLE_DEVICES=$gpus python3 valid.py --config config/{DATASET}Prompt${BACKBONE}.config \
    --gpu $gpus \
    --checkpoint model/${DATASET}Prompt${BACKBONE} \

Main Arguments:

• --config: Utilize the same configurations as PT.
• --gpu: Assign gpu ID.
• --checkpoint: Trained prompts for validation.

Cross-task Transfer

Perform cross-task transfer experiments:

bash valid_cross_task.sh

Example:

gpus=0
BASEMODEL="T5Large"
DATASET=IMDBPrompt 
PROMPT=laptopPrompt 

echo "==========================="
echo Model: config/${DATASET}${BASEMODEL}.config
echo Prompt-emb: task_prompt_emb/${PROMPT}${BASEMODEL}
echo "==========================="

CUDA_VISIBLE_DEVICES=$gpus python3 valid.py --config config/${DATASET}${BASEMODEL}.config \
    --gpu $gpus \
    --checkpoint model/${DATASET}${BASEMODEL} \
    --replacing_prompt task_prompt_emb/${PROMPT}${BASEMODEL}

Main Arguments:

• --config: Utilize the same configurations as PT.
• --gpu: Assign gpu ID.
• --replacing_prompt: Choose the source trained prompts to transfer to another tasks.

Cross-model Transfer (Train)

Train projectors for cross-model transfer:

bash train_cross_model.sh

Example:

DATASET="laptop"
MODEL_PROMPT="T5-base"
SOURCE_MODEL="T5"
TARGET_MODEL="Roberta"
source_model="t5"
target_model="roberta"

CUDA_VISIBLE_DEVICES=$gpus python3 train_cross.py --config config/crossPrompt${TARGET_MODEL}_${DATASET}_100_${source_model}_to_${target_model}.config \
    --gpu $gpus \
    --model_prompt $MODEL_PROMPT

Main Arguments:

• --config: Configurations for cross-model projectors.
• --gpu: Assign gpu ID.
• --model_prompt: Trained prompts for training cross-model projectors.

Transferability Indicators (Neuron Stimulation)

Capture values of neurons (the output values between 1st and 2nd layers of feed-forward network FFN) in every layer of a PLM [Refer to Section 6.1 in the paper]:

bash activate_neuron.sh

Example:

BACKBONE_MODEL="Roberta"

for MODEL in IMDBPrompt${BACKBONE_MODEL} laptopPrompt${BACKBONE_MODEL} MNLIPrompt${BACKBONE_MODEL} QNLIPrompt${BACKBONE_MODEL} QQPPrompt${BACKBONE_MODEL} restaurantPrompt${BACKBONE_MODEL} SST2Prompt${BACKBONE_MODEL} snliPrompt${BACKBONE_MODEL} tweetevalsentimentPrompt${BACKBONE_MODEL} movierationalesPrompt${BACKBONE_MODEL} recastnerPrompt${BACKBONE_MODEL} ethicsdeontologyPrompt${BACKBONE_MODEL} ethicsjusticePrompt${BACKBONE_MODEL} MRPCPrompt${BACKBONE_MODEL}
do
    echo "==========================="
    echo Activate_neuronPrompt${BACKBONE_MODEL}
    echo Stimulate neurons with task_prompt_emb/$MODEL
    echo "==========================="

    CUDA_VISIBLE_DEVICES=$gpus python3 activate_neuron_${BACKBONE_MODEL}.py --config config/activate_neuronPrompt${BACKBONE_MODEL}.config \
        --gpu $gpus \
        --replacing_prompt task_prompt_emb/$MODEL \
        --activate_neuron
done

Main Arguments:

• --config: Configurations for neuron stimulation (capture neuron values).
• --gpu: Assign gpu ID.
• --replacing_prompt: Choose trained prompts to stimulate neurons.
• --activate_neuron: Neuron-capture mode.

Transferability Indicators (Different Metric Results)

Utilize different metrics to evaluate transferability indicators:

python3 caculate_metric_value_for_correlation.py

Activated Neurons Demo

Colab

Citations

Please cite our paper if it is helpful to your work!

@inproceedings{su-etal-2022-transferability,
    title = "On Transferability of Prompt Tuning for Natural Language Processing",
    author = "Su, Yusheng  and
      Wang, Xiaozhi  and
      Qin, Yujia  and
      Chan, Chi-Min  and
      Lin, Yankai  and
      Wang, Huadong  and
      Wen, Kaiyue  and
      Liu, Zhiyuan  and
      Li, Peng  and
      Li, Juanzi  and
      Hou, Lei  and
      Sun, Maosong  and
      Zhou, Jie",
    booktitle = "Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jul,
    year = "2022",
    address = "Seattle, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.naacl-main.290",
    doi = "10.18653/v1/2022.naacl-main.290",
    pages = "3949--3969"
}

Contact

Yusheng Su

Mail: yushengsu.thu@gmail.com; suys19@mauls.tsinghua.edu.cn