Self-Data Distillation for Recovering Quality in Pruned Large Language Models

• URL: http://arxiv.org/abs/2410.09982v3
• Date: Tue, 12 Nov 2024 04:20:00 GMT
• Title: Self-Data Distillation for Recovering Quality in Pruned Large Language Models
• Authors: Vithursan Thangarasa, Ganesh Venkatesh, Mike Lasby, Nish Sinnadurai, Sean Lie,
• Abstract summary: One-shot pruning results in significant quality degradation, particularly in tasks requiring multi-step reasoning. To recover lost quality, supervised fine-tuning (SFT) is commonly applied, but it can lead to catastrophic forgetting. In this work, we utilize self-data distilled fine-tuning to address these challenges.
• Score: 1.5665059604715017
• License:
• Abstract: Large language models have driven significant progress in natural language processing, but their deployment requires substantial compute and memory resources. As models scale, compression techniques become essential for balancing model quality with computational efficiency. Structured pruning, which removes less critical components of the model, is a promising strategy for reducing complexity. However, one-shot pruning often results in significant quality degradation, particularly in tasks requiring multi-step reasoning. To recover lost quality, supervised fine-tuning (SFT) is commonly applied, but it can lead to catastrophic forgetting by shifting the model's learned data distribution. Therefore, addressing the degradation from both pruning and SFT is essential to preserve the original model's quality. In this work, we utilize self-data distilled fine-tuning to address these challenges. Our approach leverages the original, unpruned model to generate a distilled dataset that preserves semantic richness and mitigates catastrophic forgetting by maintaining alignment with the base model's knowledge. Empirically, we demonstrate that self-data distillation consistently outperforms standard SFT, improving average accuracy by up to 8% on the HuggingFace OpenLLM Leaderboard v1. Specifically, when pruning six decoder blocks on Llama3.1-8B Instruct (i.e., 32 to 26 layers, reducing the model size from 8.03B to 6.72B parameters), our method retains 91.2% of the original model's accuracy compared to 81.7% with SFT, while reducing real-world FLOPs by 16.3%. Furthermore, combining self-data distilled models through model merging yields enhanced quality retention. Additionally, leveraging these pruned models in speculative decoding increases token acceptance rates, thereby improving inference efficiency in applied settings.

Related papers

• Comb, Prune, Distill: Towards Unified Pruning for Vision Model Compression [24.119415458653616]
  We propose a novel unified pruning framework Comb, Prune, Distill (CPD) to address both model-agnostic and task-agnostic concerns simultaneously. Our framework employs a combing step to resolve hierarchical layer-wise dependency issues, enabling architecture independence. In image classification we achieve a speedup of up to x4.3 with a accuracy loss of 1.8% and in semantic segmentation up to x1.89 with a 5.1% loss in mIoU.
  arXiv  Detail & Related papers  (2024-08-06T09:02:31Z)
• PUMA: margin-based data pruning [51.12154122266251]
  We focus on data pruning, where some training samples are removed based on the distance to the model classification boundary (i.e., margin) We propose PUMA, a new data pruning strategy that computes the margin using DeepFool. We show that PUMA can be used on top of the current state-of-the-art methodology in robustness, and it is able to significantly improve the model performance unlike the existing data pruning strategies.
  arXiv  Detail & Related papers  (2024-05-10T08:02:20Z)
• Three-Stage Adjusted Regression Forecasting (TSARF) for Software Defect Prediction [5.826476252191368]
  Nonhomogeneous Poisson process (NHPP) SRGM are the most commonly employed models. Increased model complexity presents a challenge in identifying robust and computationally efficient algorithms.
  arXiv  Detail & Related papers  (2024-01-31T02:19:35Z)
• EsaCL: Efficient Continual Learning of Sparse Models [10.227171407348326]
  Key challenge in the continual learning setting is to efficiently learn a sequence of tasks without forgetting how to perform previously learned tasks. We propose a new method for efficient continual learning of sparse models (EsaCL) that can automatically prune redundant parameters without adversely impacting the model's predictive power.
  arXiv  Detail & Related papers  (2024-01-11T04:59:44Z)
• A-SDM: Accelerating Stable Diffusion through Redundancy Removal and Performance Optimization [54.113083217869516]
  In this work, we first explore the computational redundancy part of the network. We then prune the redundancy blocks of the model and maintain the network performance. Thirdly, we propose a global-regional interactive (GRI) attention to speed up the computationally intensive attention part.
  arXiv  Detail & Related papers  (2023-12-24T15:37:47Z)
• Accurate and Structured Pruning for Efficient Automatic Speech Recognition [23.897482741744117]
  We propose a novel compression strategy to reduce the model size and inference cost of the Conformer model. Our method achieves a 50% reduction in model size and a 28% reduction in inference cost with minimal performance loss.
  arXiv  Detail & Related papers  (2023-05-31T04:31:16Z)
• Towards Efficient Task-Driven Model Reprogramming with Foundation Models [52.411508216448716]
  Vision foundation models exhibit impressive power, benefiting from the extremely large model capacity and broad training data. However, in practice, downstream scenarios may only support a small model due to the limited computational resources or efficiency considerations. This brings a critical challenge for the real-world application of foundation models: one has to transfer the knowledge of a foundation model to the downstream task.
  arXiv  Detail & Related papers  (2023-04-05T07:28:33Z)
• Gradient-Free Structured Pruning with Unlabeled Data [57.999191898036706]
  We propose a gradient-free structured pruning framework that uses only unlabeled data. Up to 40% of the original FLOP count can be reduced with less than a 4% accuracy loss across all tasks considered.
  arXiv  Detail & Related papers  (2023-03-07T19:12:31Z)
• Deep learning model compression using network sensitivity and gradients [3.52359746858894]
  We present model compression algorithms for both non-retraining and retraining conditions. In the first case, we propose the Bin & Quant algorithm for compression of the deep learning models using the sensitivity of the network parameters. In the second case, we propose our novel gradient-weighted k-means clustering algorithm (GWK)
  arXiv  Detail & Related papers  (2022-10-11T03:02:40Z)
• Effective Model Sparsification by Scheduled Grow-and-Prune Methods [73.03533268740605]
  We propose a novel scheduled grow-and-prune (GaP) methodology without pre-training the dense models. Experiments have shown that such models can match or beat the quality of highly optimized dense models at 80% sparsity on a variety of tasks.
  arXiv  Detail & Related papers  (2021-06-18T01:03:13Z)
• Towards Practical Lipreading with Distilled and Efficient Models [57.41253104365274]
  Lipreading has witnessed a lot of progress due to the resurgence of neural networks. Recent works have placed emphasis on aspects such as improving performance by finding the optimal architecture or improving generalization. There is still a significant gap between the current methodologies and the requirements for an effective deployment of lipreading in practical scenarios. We propose a series of innovations that significantly bridge that gap: first, we raise the state-of-the-art performance by a wide margin on LRW and LRW-1000 to 88.5% and 46.6%, respectively using self-distillation.
  arXiv  Detail & Related papers  (2020-07-13T16:56:27Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.