Abstract:We introduce compression laws for language language models (LLMs). While recent scaling laws have sought to understand how LLMs scale with respect to model size, pre-training data, and computational resources, we focus on understanding how model compression affects the performance of a pre-trained LLM on downstream tasks. We empirically examine the effects of structured model compression on LLMs through over $1000$ experiments across eight models with sizes ranging from $0.5B$ to $14B$ parameters. Our findings indicate that the test cross-entropy loss increases quadratically with the compression ratio, whereas performance on downstream tasks declines only linearly. Our study emphasizes the importance of recovery fine-tuning in enhancing generation loss, showing that the test loss of compressed LLMs can improve by up to 55% with recovery fine-tuning. At higher compression ratios (up to 90%), compressed LLMs demonstrate a speed increase of 60% during inference compared to their uncompressed counterparts, compensating for the performance degradation at this level. However, for smaller models ($\le 7B$), the computational gains are limited, peaking at just 35%. We conclude that model compression can be highly beneficial for larger models, especially when a smaller model within the same computational budget is not available. These insights provide the practical guidelines for utilizing model compression techniques for adopting LLMs in real-life applications in resource-constrained settings.
Abstract:The ever-increasing size of large language models (LLMs) presents significant challenges for deployment due to their heavy computational and memory requirements. Current model pruning techniques attempt to alleviate these issues by relying heavily on external calibration datasets to determine which parameters to prune or compress, thus limiting their flexibility and scalability across different compression ratios. Moreover, these methods often cause severe performance degradation, particularly in downstream tasks, when subjected to higher compression rates. In this paper, we propose PruneNet, a novel model compression method that addresses these limitations by reformulating model pruning as a policy learning process. PruneNet decouples the pruning process from the model architecture, eliminating the need for calibration datasets. It learns a stochastic pruning policy to assess parameter importance solely based on intrinsic model properties while preserving the spectral structure to minimize information loss. PruneNet can compress the LLaMA-2-7B model in just 15 minutes, achieving over 80% retention of its zero-shot performance with a 30% compression ratio, outperforming existing methods that retain only 75% performance. Furthermore, on complex multitask language understanding tasks, PruneNet demonstrates its robustness by preserving up to 80% performance of the original model, proving itself a superior alternative to conventional structured compression techniques.
Abstract:Contextual bandit algorithms are at the core of many applications, including recommender systems, clinical trials, and optimal portfolio selection. One of the most popular problems studied in the contextual bandit literature is to maximize the sum of the rewards in each round by ensuring a sublinear regret against the best-fixed context-dependent policy. However, in many applications, the cumulative reward is not the right objective - the bandit algorithm must be fair in order to avoid the echo-chamber effect and comply with the regulatory requirements. In this paper, we consider the $\alpha$-Fair Contextual Bandits problem, where the objective is to maximize the global $\alpha$-fair utility function - a non-decreasing concave function of the cumulative rewards in the adversarial setting. The problem is challenging due to the non-separability of the objective across rounds. We design an efficient algorithm that guarantees an approximately sublinear regret in the full-information and bandit feedback settings.