Junk DNA Hypothesis: Pruning Small Pre-Trained Weights $\textitIrreversibly$ and $\textitMonotonically$ Impairs “Difficult" Downstream Tasks in LLMs

Lu Yin, Ajay Kumar Jaiswal, Shiwei Liu, Souvik Kundu, Zhangyang Wang
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:57053-57068, 2024.

Abstract

We present Junk DNA Hypothesis by adopting a novel task-centric angle for the pre-trained weights of large language models (LLMs). It has been believed that weights in LLMs contain significant redundancy, leading to the conception that a considerable chunk of the parameters can be removed by pruning without compromising performance. Contrary to this belief, this paper presents a counter-argument: small-magnitude weights of pre-trained model weights encode vital knowledge essential for tackling difficult downstream tasks - manifested as the monotonic relationship between the performance drop of downstream tasks across the difficulty spectrum, as we prune more pre-trained weights by magnitude. Moreover, we reveal that these seemingly inconsequential weights can result in irreparable loss of knowledge and performance degradation in difficult tasks, even when downstream continual training is allowed. Interestingly, our evaluations show that the other popular compression, namely quantization fail to exhibit similar “monotonic" effect and does not as convincingly disentangle this task-difficulty information. To study formally, we introduce several quantifiable metrics to gauge the downstream task difficulty: (a) within the same task category, and (b) across different task categories. Our extensive experiments substantiate the Junk DNA Hypothesis across a diverse range of model sizes, tasks, datasets, and even pruning methods. Codes are available at https://github.com/VITA-Group/Junk_DNA_Hypothesis.git.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-yin24b, title = {Junk {DNA} Hypothesis: Pruning Small Pre-Trained Weights $\textit{Irreversibly}$ and $\textit{Monotonically}$ Impairs “Difficult" Downstream Tasks in {LLM}s}, author = {Yin, Lu and Jaiswal, Ajay Kumar and Liu, Shiwei and Kundu, Souvik and Wang, Zhangyang}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {57053--57068}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/yin24b/yin24b.pdf}, url = {https://proceedings.mlr.press/v235/yin24b.html}, abstract = {We present Junk DNA Hypothesis by adopting a novel task-centric angle for the pre-trained weights of large language models (LLMs). It has been believed that weights in LLMs contain significant redundancy, leading to the conception that a considerable chunk of the parameters can be removed by pruning without compromising performance. Contrary to this belief, this paper presents a counter-argument: small-magnitude weights of pre-trained model weights encode vital knowledge essential for tackling difficult downstream tasks - manifested as the monotonic relationship between the performance drop of downstream tasks across the difficulty spectrum, as we prune more pre-trained weights by magnitude. Moreover, we reveal that these seemingly inconsequential weights can result in irreparable loss of knowledge and performance degradation in difficult tasks, even when downstream continual training is allowed. Interestingly, our evaluations show that the other popular compression, namely quantization fail to exhibit similar “monotonic" effect and does not as convincingly disentangle this task-difficulty information. To study formally, we introduce several quantifiable metrics to gauge the downstream task difficulty: (a) within the same task category, and (b) across different task categories. Our extensive experiments substantiate the Junk DNA Hypothesis across a diverse range of model sizes, tasks, datasets, and even pruning methods. Codes are available at https://github.com/VITA-Group/Junk_DNA_Hypothesis.git.} }
Endnote
%0 Conference Paper %T Junk DNA Hypothesis: Pruning Small Pre-Trained Weights $\textitIrreversibly$ and $\textitMonotonically$ Impairs “Difficult" Downstream Tasks in LLMs %A Lu Yin %A Ajay Kumar Jaiswal %A Shiwei Liu %A Souvik Kundu %A Zhangyang Wang %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-yin24b %I PMLR %P 57053--57068 %U https://proceedings.mlr.press/v235/yin24b.html %V 235 %X We present Junk DNA Hypothesis by adopting a novel task-centric angle for the pre-trained weights of large language models (LLMs). It has been believed that weights in LLMs contain significant redundancy, leading to the conception that a considerable chunk of the parameters can be removed by pruning without compromising performance. Contrary to this belief, this paper presents a counter-argument: small-magnitude weights of pre-trained model weights encode vital knowledge essential for tackling difficult downstream tasks - manifested as the monotonic relationship between the performance drop of downstream tasks across the difficulty spectrum, as we prune more pre-trained weights by magnitude. Moreover, we reveal that these seemingly inconsequential weights can result in irreparable loss of knowledge and performance degradation in difficult tasks, even when downstream continual training is allowed. Interestingly, our evaluations show that the other popular compression, namely quantization fail to exhibit similar “monotonic" effect and does not as convincingly disentangle this task-difficulty information. To study formally, we introduce several quantifiable metrics to gauge the downstream task difficulty: (a) within the same task category, and (b) across different task categories. Our extensive experiments substantiate the Junk DNA Hypothesis across a diverse range of model sizes, tasks, datasets, and even pruning methods. Codes are available at https://github.com/VITA-Group/Junk_DNA_Hypothesis.git.
APA
Yin, L., Jaiswal, A.K., Liu, S., Kundu, S. & Wang, Z.. (2024). Junk DNA Hypothesis: Pruning Small Pre-Trained Weights $\textitIrreversibly$ and $\textitMonotonically$ Impairs “Difficult" Downstream Tasks in LLMs. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:57053-57068 Available from https://proceedings.mlr.press/v235/yin24b.html.

Related Material