# Do Compressed LLMs Forget Knowledge? An Experimental Study with Practical Implications

arXiv: https://arxiv.org/abs/2310.00867 · HTML: https://mlsanity.com/paper/2310.00867/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Builds on 18 of its 22 references that are in the map.
- First posted on arXiv 2 Oct 2023.

## Builds on (most cited first, 18 of 18)

- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://mlsanity.com/paper/2101.00190/) (2021)
- [The Power of Scale for Parameter-Efficient Prompt Tuning](https://mlsanity.com/paper/2104.08691/) (2021)
- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension](https://mlsanity.com/paper/1705.03551/) (2017)
- [PIQA: Reasoning about Physical Commonsense in Natural Language](https://mlsanity.com/paper/1911.11641/) (2019)
- [Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding](https://mlsanity.com/paper/1510.00149/) (2015)
- [GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers](https://mlsanity.com/paper/2210.17323/) (2022)
- [SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models](https://mlsanity.com/paper/2211.10438/) (2022)
- [SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot](https://mlsanity.com/paper/2301.00774/) (2023)
- [FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance](https://mlsanity.com/paper/2305.05176/) (2023)
- [Crowdsourcing Multiple Choice Science Questions](https://mlsanity.com/paper/1707.06209/) (2017)
- [ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers](https://mlsanity.com/paper/2206.01861/) (2022)
- [Compress, Then Prompt: Improving Accuracy-Efficiency Trade-off of LLM Inference with Transferable Prompt](https://mlsanity.com/paper/2305.11186/) (2023)
- [Model ensemble instead of prompt fusion: a sample-specific knowledge transfer method for few-shot prompt tuning](https://mlsanity.com/paper/2210.12587/) (2022)

## Built on by (most cited first, 6 of 6)

- [REPA Works Until It Doesn't: Early-Stopped, Holistic Alignment Supercharges Diffusion Training](https://mlsanity.com/paper/2505.16792/) (2025)
- [Investigating Social Bias Changes in Quantized Language Models](https://mlsanity.com/paper/2602.06181/) (2026)
- [Composable Interventions for Language Models](https://mlsanity.com/paper/2407.06483/) (2024)
- [Missing vs. Unused Knowledge Hypothesis for Language Model Bottlenecks in Patent Understanding](https://mlsanity.com/paper/2505.12452/) (2025)
- [Polysemy of Synthetic Neurons Towards a New Type of Explanatory Categorical Vector Spaces](https://mlsanity.com/paper/2505.07831/) (2025)
- [Rethinking Calibration for High-Sparsity LLM Pruning: From Single Sources to Capability Coverage](https://mlsanity.com/paper/2606.03328/) (2026)

## Abstract (arXiv)

Compressing Large Language Models (LLMs) often leads to reduced performance, especially for knowledge-intensive tasks. In this work, we dive into how compression damages LLMs' inherent knowledge and the possible remedies. We start by proposing two conjectures on the nature of the damage: one is certain knowledge being forgotten (or erased) after LLM compression, hence necessitating the compressed model to (re)learn from data with additional parameters; the other presumes that knowledge is internally displaced and hence one requires merely "inference re-direction" with input-side augmentation such as prompting, to recover the knowledge-related performance. Extensive experiments are then designed to (in)validate the two conjectures. We observe the promise of prompting in comparison to model tuning; we further unlock prompting's potential by introducing a variant called Inference-time Dynamic Prompting (IDP), that can effectively increase prompt diversity without incurring any inference overhead. Our experiments consistently suggest that compared to the classical re-training alternatives such as LoRA, prompting with IDP leads to better or comparable post-compression performance recovery, while saving the extra parameter size by 21x and reducing inference latency by 60%. Our experiments hence strongly endorse the conjecture of "knowledge displaced" over "knowledge forgotten", and shed light on a new efficient mechanism to restore compressed LLM performance. We additionally visualize and analyze the different attention and activation patterns between prompted and re-trained models, demonstrating they achieve performance recovery in two different regimes.