Parameter-Efficient Fine-Tuning Design Spaces

Authors: Jiaao Chen, Aston Zhang, Xingjian Shi, Mu Li, Alex Smola, Diyi Yang

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments reveal the following design patterns... These patterns lead to new methods for parameter-efficient fine-tuning, which we show experimentally outperform existing strategies across various backbone models and NLP tasks.
Researcher Affiliation Collaboration Georgia Institute of Technology, Amazon Web Services, Stanford University
Pseudocode No The paper describes its methods and components in detail but does not include any explicitly labeled pseudocode blocks or algorithms in a structured format.
Open Source Code Yes We will release our code at https://github.com/amazon-science/peft-design-spaces.
Open Datasets Yes Our process is based on the average performance on the widely-used GLUE benchmark (Wang et al., 2018). It covers a wide range of natural language understanding tasks.
Dataset Splits Yes Our process is based on the average performance on the widely-used GLUE benchmark (Wang et al., 2018). ... To quantify the overall quality of models in any design space Si with a low-compute, low-epoch regime (Radosavovic et al., 2020), we randomly sample 100 models from Si, fine-tune with only 3 epochs.
Hardware Specification Yes All the experiments were performed using 8 A100 GPUs.
Software Dependencies No The paper mentions 'We use Hugging Face Transformers for our implementations' but does not specify a version number for this or any other software dependency.
Experiment Setup Yes The batch size was 128 for base models and 64 for large models. The maximum learning rate was 5e-5 and the maximum number of training epochs was set to be either 5 or 10.