Sparse Finetuning for Inference Acceleration of Large Language Models

We consider the problem of accurate sparse finetuning of large language models (LLMs), that is, finetuning pretrained LLMs on specialized tasks, while inducing sparsity in their weights. On the accuracy side, we observe that standard loss-based finetuning may fail to recover accuracy, especially at high sparsities. To address this, we perform a detailed study of distillation-type losses, determining an L2-based distillation approach we term SquareHead which enables accurate recovery even at higher sparsities, across all model types. On the practical efficiency side, we show that sparse LLMs can be executed with speedups by taking advantage of sparsity, for both CPU and GPU runtimes. While the standard approach is to leverage sparsity for computational reduction, we observe that in the case of memory-bound LLMs sparsity can also be leveraged for reducing memory bandwidth. We exhibit end-to-end results showing speedups due to sparsity, while recovering accuracy, on T5 (language translation), Whisper (speech translation), and open GPT-type (MPT for text generation). For MPT text generation, we show for the first time that sparse finetuning can reach 75% sparsity without accuracy drops, provide notable end-to-end speedups for both CPU and GPU inference, and highlight that sparsity is also compatible with quantization approaches. Models and software for reproducing our results are provided in Section 6.

SPADE: Sparsity-Guided Debugging for Deep Neural Networks

Interpretability, broadly defined as mechanisms for understanding why and how machine learning models reach their decisions, is one of the key open goals at the intersection of deep learning theory and practice. Towards this goal, multiple tools have been proposed to aid a human examiner in reasoning about a network's behavior in general or on a set of instances. However, the outputs of these tools-such as input saliency maps or neuron visualizations-are frequently difficult for a human to interpret, or even misleading, due, in particular, to the fact that neurons can be multifaceted, i.e., a single neuron can be associated with multiple distinct feature combinations. In this paper, we present a new general approach to address this problem, called SPADE, which, given a trained model and a target sample, uses sample-targeted pruning to provide a "trace" of the network's execution on the sample, reducing the network to the connections that are most relevant to the specific prediction. We demonstrate that preprocessing with SPADE significantly increases both the accuracy of image saliency maps across several interpretability methods and the usefulness of neuron visualizations, aiding humans in reasoning about network behavior. Our findings show that sample-specific pruning of connections can disentangle multifaceted neurons, leading to consistently improved interpretability.

Scaling Laws for Sparsely-Connected Foundation Models

We explore the impact of parameter sparsity on the scaling behavior of Transformers trained on massive datasets (i.e., "foundation models"), in both vision and language domains. In this setting, we identify the first scaling law describing the relationship between weight sparsity, number of non-zero parameters, and amount of training data, which we validate empirically across model and data scales; on ViT/JFT-4B and T5/C4. These results allow us to characterize the "optimal sparsity", the sparsity level which yields the best performance for a given effective model size and training budget. For a fixed number of non-zero parameters, we identify that the optimal sparsity increases with the amount of data used for training. We also extend our study to different sparsity structures (such as the hardware-friendly n:m pattern) and strategies (such as starting from a pretrained dense model). Our findings shed light on the power and limitations of weight sparsity across various parameter and computational settings, offering both theoretical understanding and practical implications for leveraging sparsity towards computational efficiency improvements.

Accurate Neural Network Pruning Requires Rethinking Sparse Optimization

Obtaining versions of deep neural networks that are both highly-accurate and highly-sparse is one of the main challenges in the area of model compression, and several high-performance pruning techniques have been investigated by the community. Yet, much less is known about the interaction between sparsity and the standard stochastic optimization techniques used for training sparse networks, and most existing work uses standard dense schedules and hyperparameters for training sparse networks. In this work, we examine the impact of high sparsity on model training using the standard computer vision and natural language processing sparsity benchmarks. We begin by showing that using standard dense training recipes for sparse training is suboptimal, and results in under-training. We provide new approaches for mitigating this issue for both sparse pre-training of vision models (e.g. ResNet50/ImageNet) and sparse fine-tuning of language models (e.g. BERT/GLUE), achieving state-of-the-art results in both settings in the high-sparsity regime, and providing detailed analyses for the difficulty of sparse training in both scenarios. Our work sets a new threshold in terms of the accuracies that can be achieved under high sparsity, and should inspire further research into improving sparse model training, to reach higher accuracies under high sparsity, but also to do so efficiently.

QIGen: Generating Efficient Kernels for Quantized Inference on Large Language Models

We present ongoing work on a new automatic code generation approach for supporting quantized generative inference on LLMs such as LLaMA or OPT on off-the-shelf CPUs. Our approach is informed by the target architecture and a performance model, including both hardware characteristics and method-specific accuracy constraints. Results on CPU-based inference for LLaMA models show that our approach can lead to high performance and high accuracy, comparing favorably to the best existing open-source solution. A preliminary implementation is available at https://github.com/IST-DASLab/QIGen.

Error Feedback Can Accurately Compress Preconditioners

Leveraging second-order information at the scale of deep networks is one of the main lines of approach for improving the performance of current optimizers for deep learning. Yet, existing approaches for accurate full-matrix preconditioning, such as Full-Matrix Adagrad (GGT) or Matrix-Free Approximate Curvature (M-FAC) suffer from massive storage costs when applied even to medium-scale models, as they must store a sliding window of gradients, whose memory requirements are multiplicative in the model dimension. In this paper, we address this issue via an efficient and simple-to-implement error-feedback technique that can be applied to compress preconditioners by up to two orders of magnitude in practice, without loss of convergence. Specifically, our approach compresses the gradient information via sparsification or low-rank compression \emph{before} it is fed into the preconditioner, feeding the compression error back into future iterations. Extensive experiments on deep neural networks for vision show that this approach can compress full-matrix preconditioners by up to two orders of magnitude without impact on accuracy, effectively removing the memory overhead of full-matrix preconditioning for implementations of full-matrix Adagrad (GGT) and natural gradient (M-FAC). Our code is available at https://github.com/IST-DASLab/EFCP.

Decentralized Learning Dynamics in the Gossip Model

We study a distributed multi-armed bandit setting among a population of $n$ memory-constrained nodes in the gossip model: at each round, every node locally adopts one of $m$ arms, observes a reward drawn from the arm's (adversarially chosen) distribution, and then communicates with a randomly sampled neighbor, exchanging information to determine its policy in the next round. We introduce and analyze several families of dynamics for this task that are decentralized: each node's decision is entirely local and depends only on its most recently obtained reward and that of the neighbor it sampled. We show a connection between the global evolution of these decentralized dynamics with a certain class of "zero-sum" multiplicative weight update algorithms, and we develop a general framework for analyzing the population-level regret of these natural protocols. Using this framework, we derive sublinear regret bounds under a wide range of parameter regimes (i.e., the size of the population and number of arms) for both the stationary reward setting (where the mean of each arm's distribution is fixed over time) and the adversarial reward setting (where means can vary over time). Further, we show that these protocols can approximately optimize convex functions over the simplex when the reward distributions are generated from a stochastic gradient oracle.

SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression

Recent advances in large language model (LLM) pretraining have led to high-quality LLMs with impressive abilities. By compressing such LLMs via quantization to 3-4 bits per parameter, they can fit into memory-limited devices such as laptops and mobile phones, enabling personalized use. However, quantization down to 3-4 bits per parameter usually leads to moderate-to-high accuracy losses, especially for smaller models in the 1-10B parameter range, which are well-suited for edge deployments. To address this accuracy issue, we introduce the Sparse-Quantized Representation (SpQR), a new compressed format and quantization technique which enables for the first time near-lossless compression of LLMs across model scales, while reaching similar compression levels to previous methods. SpQR works by identifying and isolating outlier weights, which cause particularly-large quantization errors, and storing them in higher precision, while compressing all other weights to 3-4 bits, and achieves relative accuracy losses of less than 1% in perplexity for highly-accurate LLaMA and Falcon LLMs. This makes it possible to run 33B parameter LLM on a single 24 GB consumer GPU without any performance degradation at 15% speedup thus making powerful LLMs available to consumer without any downsides. SpQR comes with efficient algorithms for both encoding weights into its format, as well as decoding them efficiently at runtime. Specifically, we provide an efficient GPU inference algorithm for SpQR which yields faster inference than 16-bit baselines at similar accuracy, while enabling memory compression gains of more than 4x.