EvoPress: Towards Optimal Dynamic Model Compression via Evolutionary Search

• URL: http://arxiv.org/abs/2410.14649v1
• Date: Fri, 18 Oct 2024 17:46:37 GMT
• Title: EvoPress: Towards Optimal Dynamic Model Compression via Evolutionary Search
• Authors: Oliver Sieberling, Denis Kuznedelev, Eldar Kurtic, Dan Alistarh,
• Abstract summary: We propose a new and general approach for dynamic compression that is provably optimal in a given input range. We show that these theoretical guarantees lead to highly competitive practical performance for dynamic compression of Llama, Mistral and Phi models.
• Score: 33.86918407429272
• License:
• Abstract: The high computational costs of large language models (LLMs) have led to a flurry of research on LLM compression, via methods such as quantization, sparsification, or structured pruning. A new frontier in this area is given by \emph{dynamic, non-uniform} compression methods, which adjust the compression levels (e.g., sparsity) per-block or even per-layer in order to minimize accuracy loss, while guaranteeing a global compression threshold. Yet, current methods rely on heuristics for identifying the "importance" of a given layer towards the loss, based on assumptions such as \emph{error monotonicity}, i.e. that the end-to-end model compression error is proportional to the sum of layer-wise errors. In this paper, we revisit this area, and propose a new and general approach for dynamic compression that is provably optimal in a given input range. We begin from the motivating observation that, in general, \emph{error monotonicity does not hold for LLMs}: compressed models with lower sum of per-layer errors can perform \emph{worse} than models with higher error sums. To address this, we propose a new general evolutionary framework for dynamic LLM compression called EvoPress, which has provable convergence, and low sample and evaluation complexity. We show that these theoretical guarantees lead to highly competitive practical performance for dynamic compression of Llama, Mistral and Phi models. Via EvoPress, we set new state-of-the-art results across all compression approaches: structural pruning (block/layer dropping), unstructured sparsity, as well as quantization with dynamic bitwidths. Our code is available at https://github.com/IST-DASLab/EvoPress.

Related papers

• LoRC: Low-Rank Compression for LLMs KV Cache with a Progressive Compression Strategy [59.1298692559785]
  Key-Value ( KV) cache is crucial component in serving transformer-based autoregressive large language models (LLMs) Existing approaches to mitigate this issue include: (1) efficient attention variants integrated in upcycling stages; (2) KV cache compression at test time; and (3) KV cache compression at test time. We propose a low-rank approximation of KV weight matrices, allowing plug-in integration with existing transformer-based LLMs without model retraining. Our method is designed to function without model tuning in upcycling stages or task-specific profiling in test stages.
  arXiv  Detail & Related papers  (2024-10-04T03:10:53Z)
• Fundamental Limits of Prompt Compression: A Rate-Distortion Framework for Black-Box Language Models [21.025001473355996]
  We formalize the problem of prompt compression for large language models (LLMs) We present a framework to unify token-level prompt compression methods which create hard prompts for black-box models. We show that there is a large gap between the performance of current prompt compression methods and the optimal strategy.
  arXiv  Detail & Related papers  (2024-07-22T09:40:13Z)
• Compression of Structured Data with Autoencoders: Provable Benefit of Nonlinearities and Depth [83.15263499262824]
  We prove that gradient descent converges to a solution that completely disregards the sparse structure of the input. We show how to improve upon Gaussian performance for the compression of sparse data by adding a denoising function to a shallow architecture. We validate our findings on image datasets, such as CIFAR-10 and MNIST.
  arXiv  Detail & Related papers  (2024-02-07T16:32:29Z)
• Activations and Gradients Compression for Model-Parallel Training [85.99744701008802]
  We study how simultaneous compression of activations and gradients in model-parallel distributed training setup affects convergence. We find that gradients require milder compression rates than activations. Experiments also show that models trained with TopK perform well only when compression is also applied during inference.
  arXiv  Detail & Related papers  (2024-01-15T15:54:54Z)
• Just CHOP: Embarrassingly Simple LLM Compression [27.64461490974072]
  Large language models (LLMs) enable unparalleled few- and zero-shot reasoning capabilities but at a high computational footprint. We show that simple layer pruning coupled with an extended language model pretraining produces state-of-the-art results against structured and even semi-structured compression of models at a 7B scale. We also show how distillation, which has been super effective in task-agnostic compression of smaller BERT-style models, becomes inefficient against our simple pruning technique.
  arXiv  Detail & Related papers  (2023-05-24T08:18:35Z)
• OPQ: Compressing Deep Neural Networks with One-shot Pruning-Quantization [32.60139548889592]
  We propose a novel One-shot Pruning-Quantization (OPQ) in this paper. OPQ analytically solves the compression allocation with pre-trained weight parameters only. We propose a unified channel-wise quantization method that enforces all channels of each layer to share a common codebook.
  arXiv  Detail & Related papers  (2022-05-23T09:05:25Z)
• Compression of Generative Pre-trained Language Models via Quantization [62.80110048377957]
  We find that previous quantization methods fail on generative tasks due to the textithomogeneous word embeddings We propose a token-level contrastive distillation to learn distinguishable word embeddings, and a module-wise dynamic scaling to make quantizers adaptive to different modules.
  arXiv  Detail & Related papers  (2022-03-21T02:11:35Z)
• Compressing Neural Networks: Towards Determining the Optimal Layer-wise Decomposition [62.41259783906452]
  We present a novel global compression framework for deep neural networks. It automatically analyzes each layer to identify the optimal per-layer compression ratio. Our results open up new avenues for future research into the global performance-size trade-offs of modern neural networks.
  arXiv  Detail & Related papers  (2021-07-23T20:01:30Z)
• Towards Compact CNNs via Collaborative Compression [166.86915086497433]
  We propose a Collaborative Compression scheme, which joints channel pruning and tensor decomposition to compress CNN models. We achieve 52.9% FLOPs reduction by removing 48.4% parameters on ResNet-50 with only a Top-1 accuracy drop of 0.56% on ImageNet 2012.
  arXiv  Detail & Related papers  (2021-05-24T12:07:38Z)
• Successive Pruning for Model Compression via Rate Distortion Theory [15.598364403631528]
  We study NN compression from an information-theoretic approach and show that rate distortion theory suggests pruning to achieve the theoretical limits of NN compression. Our derivation also provides an end-to-end compression pipeline involving a novel pruning strategy. Our method consistently outperforms the existing pruning strategies and reduces the pruned model's size by 2.5 times.
  arXiv  Detail & Related papers  (2021-02-16T18:17:57Z)
• A flexible, extensible software framework for model compression based on the LC algorithm [10.787390511207683]
  We propose a software framework that allows a user to compress a neural network or other machine learning model with minimal effort. The library is written in Python and PyTorch and available in Github.
  arXiv  Detail & Related papers  (2020-05-15T21:14:48Z)

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.