Balanced Column-Wise Block Pruning for Maximizing GPU Parallelism

Authors: Cheonjun Park, Mincheol Park, Hyun Jae Oh, Minkyu Kim, Myung Kuk Yoon, Suhyun Kim, Won Woo Ro

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate that BCBP is superior to previous pruning methods through comprehensive experiments. and Experiments Datasets and Models We evaluate the performance of BCBP using IWSLT English-Vietnamese (Luong and Manning 2016), SQu AD (Rajpurkar, Jia, and Liang 2018), and Image Net dataset (Deng et al. 2009).
Researcher Affiliation Collaboration Cheonjun Park1, Mincheol Park1,2, Hyun Jae Oh3, Minkyu Kim1, Myung Kuk Yoon4, Suhyun Kim2, and Won Woo Ro1* 1Yonsei University 2Korea Institute of Science and Technology 3Samsung Electronics 4Ewha Womans University
Pseudocode No The paper describes the process of BCBP in text but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The footnote '1https://github.com/cheonjun-park/appendix1' refers to Appendix 1 for 'additional verification of the following DNN models' but does not explicitly state that the source code for the BCBP methodology itself is provided at this link or elsewhere.
Open Datasets Yes Datasets and Models We evaluate the performance of BCBP using IWSLT English-Vietnamese (Luong and Manning 2016), SQu AD (Rajpurkar, Jia, and Liang 2018), and Image Net dataset (Deng et al. 2009).
Dataset Splits No Image Net contains 1.28M training images and 50K test images. The paper mentions training and testing sets, but does not explicitly describe a separate validation split with specific details for reproducibility.
Hardware Specification Yes We evaluate BCBP using NVIDIA RTX 2080 TI GPUs.
Software Dependencies No The paper mentions using 'Py Torch framework' but does not specify a version number or other software dependencies with their versions.
Experiment Setup Yes For settings of the fine-tuning, we use SGD optimizer with the weight decay, 1 10 4 and the momentum as 0.9. We set a batch size of 256 and a learning rate of 0.0001.