MuSKPrompt

Released code for our NAACL24 paper: Multi-Scale Prompt Memory-Augmented Model for Black-Box Scenarios

Overview of MuSKPromt. LLMs are frozen and inaccessible to internal parameters and gradient information. Knowledge is extracted using prompts of four different scales ((m = 4)), and stored in various datastores constituting the memory bank. $P_{ic}$ represents the prompt at the i-th scale, where $c$ denotes the number of examples selected from each class. The bottom-left illustrates an example of a prompt with a scale size of 1.

In this paper, we present MuSKPrompt (Multi-scale Knowledge Prompt for Memory Model), an efficient multi-scale knowledge prompt-based memory model in black-box few-shot text classification task. MuSKPrompt extracts instance-level and class-level knowledge at different scales and stores them in memory banks during training. Then, it references multi-scale memory banks to perform quick inference on new samples via a novel scoring module. MuSKPrompt achieves competitive performance in limited data through multi-scale instance-level and class-level knowledge. Moreover, it realizes gradient-free optimization with zero training parameters in the black-box scenario.

Preparation

Requirements

• pytorch == 1.13.1+cu116
• transformers == 4.20.1
• python == 3.8

Model and Data

Prepare your LLMs in ./llm/, I personally prefer download them myself and configure the local path in scripts. Download dataset from url and unzip them in ./data.

How to Run

After downloading the model and data, and checking the configuration, you can run the following script.

bash run_muti_scale.sh 

Citation

• If you are interested in our approach, feel free to cite us.
• If you find this repo useful, please cite us as:

@inproceedings{kuang-etal-2024-multi,
    title = "Multi-Scale Prompt Memory-Augmented Model for Black-Box Scenarios",
    author = "Kuang, Xiaojun  and
      Chen, C. L. Philip  and
      Li, Shuzhen  and
      Zhang, Tong",
    booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
    month = jun,
    year = "2024",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.naacl-long.98",
}

Acknowledge

Our work is based on KNNPrompting, with a similar data experiment setup and code architecture to it. Thanks to the open source code for saving us a lot of time!