How much information do NLP tasks really need from a transformer's attention mechanism at application-time (inference)? From recent work, we know that there is sparsity in transformers and that the floating-points within its computation can be discretized to fewer values with minimal loss to task accuracies. However, this requires retraining or even creating entirely new models, both of which can be expensive and carbon-emitting. Focused on optimizations that do not require training, we systematically study the full range of typical attention values necessary. This informs the design of an inference-time quantization technique using both pruning and log-scaled mapping which produces only a few (e.g. $2^3$) unique values. Over the tasks of question answering and sentiment analysis, we find nearly 80% of attention values can be pruned to zeros with minimal ($< 1.0\%$) relative loss in accuracy. We use this pruning technique in conjunction with quantizing the attention values to only a 3-bit format, without retraining, resulting in only a 0.8% accuracy reduction on question answering with fine-tuned RoBERTa.
Instituto de Telecomunicações（电信研究所）；Instituto Superior Técnico（高等技术学院）；LUMLIS (Lisbon ELLIS Unit)（LUMLIS（里斯本ELLIS单元））；Mila, Université de Montréal（蒙特利尔大学米拉研究所）；Language Technologies Institute, Carnegie Mellon University（卡内基梅隆大学语言技术研究所）；Unbabel（Unbabel公司）；Kaufland e-commerce（考夫兰电子商务）
Transformers' quadratic complexity with respect to the input sequence length has motivated a body of work on efficient sparse approximations to softmax. An alternative path, used by entmax transformers, consists of having built-in exact sparse attention; however this approach still requires quadratic computation. In this paper, we propose Sparsefinder, a simple model trained to identify the sparsity pattern of entmax attention before computing it. We experiment with three variants of our method, based on distances, quantization, and clustering, on two tasks: machine translation (attention in t…
We introduce a novel run-time method for significantly reducing the accuracy loss associated with quantizing BERT-like models to 8-bit integers. Existing methods for quantizing models either modify the training procedure,or they require an additional calibration step to adjust parameters that also requires a selected held-out dataset. Our method permits taking advantage of quantization without the need for these adjustments. We present results on several NLP tasks demonstrating the usefulness of this technique.
Attention mechanisms have become ubiquitous in NLP. Recent architectures, notably the Transformer, learn powerful context-aware word representations through layered, multi-headed attention. The multiple heads learn diverse types of word relationships. However, with standard softmax attention, all attention heads are dense, assigning a non-zero weight to all context words. In this work, we introduce the adaptively sparse Transformer, wherein attention heads have flexible, context-dependent sparsity patterns. This sparsity is accomplished by replacing softmax with $α$-entmax: a differentiable ge…
Natural Language Processing Lab., Northeastern University（东北大学自然语言处理实验室）；NiuTrans Research（小牛翻译研究院）；CAS Key Laboratory of Behavioral Science, Institute of Psychology, CAS（中国科学院心理研究所行为科学重点实验室）
8-bit integer inference, as a promising direction in reducing both the latency and storage of deep neural networks, has made great progress recently. On the other hand, previous systems still rely on 32-bit floating point for certain functions in complex models (e.g., Softmax in Transformer), and make heavy use of quantization and de-quantization. In this work, we show that after a principled modification on the Transformer architecture, dubbed Integer Transformer, an (almost) fully 8-bit integer inference algorithm Scale Propagation could be derived. De-quantization is adopted when necessary,…
Qualcomm AI Research（高通人工智能研究（Qualcomm AI Research））；Qualcomm Technologies, Inc.（高通公司（Qualcomm Technologies, Inc.））
Transformer-based architectures have become the de-facto standard models for a wide range of Natural Language Processing tasks. However, their memory footprint and high latency are prohibitive for efficient deployment and inference on resource-limited devices. In this work, we explore quantization for transformers. We show that transformers have unique quantization challenges -- namely, high dynamic activation ranges that are difficult to represent with a low bit fixed-point format. We establish that these activations contain structured outliers in the residual connections that encourage speci…
Hanyang University（汉阳大学）；Seoul National University（首尔国立大学）
Pre-trained Transformer models such as BERT have shown great success in a wide range of applications, but at the cost of substantial increases in model complexity. Quantization-aware training (QAT) is a promising method to lower the implementation cost and energy consumption. However, aggressive quantization below 2-bit causes considerable accuracy degradation due to unstable convergence, especially when the downstream dataset is not abundant. This work proposes a proactive knowledge distillation method called Teacher Intervention (TI) for fast converging QAT of ultra-low precision pre-trained…
In recent years, large pre-trained Transformer networks have demonstrated dramatic improvements in many natural language understanding tasks. However, the huge size of these models brings significant challenges to their fine-tuning and online deployment due to latency and cost constraints. New hardware supporting both N:M semi-structured sparsity and low-precision integer computation is a promising solution to boost DNN model serving efficiency. However, there have been very few studies that systematically investigate to what extent pre-trained Transformer networks benefit from the combination…
Knowledge distillation (KD) has been a ubiquitous method for model compression to strengthen the capability of a lightweight model with the transferred knowledge from the teacher. In particular, KD has been employed in quantization-aware training (QAT) of Transformer encoders like BERT to improve the accuracy of the student model with the reduced-precision weight parameters. However, little is understood about which of the various KD approaches best fits the QAT of Transformers. In this work, we provide an in-depth analysis of the mechanism of KD on attention recovery of quantized large Transf…
State-of-the-art neural language models represented by Transformers are becoming increasingly complex and expensive for practical applications. Low-bit deep neural network quantization techniques provides a powerful solution to dramatically reduce their model size. Current low-bit quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of the system to quantization errors. To this end, novel mixed precision DNN quantization methods are proposed in this paper. The optimal local precision settings are automatically learne…
The deployment of widely used Transformer architecture is challenging because of heavy computation load and memory overhead during inference, especially when the target device is limited in computational resources such as mobile or edge devices. Quantization is an effective technique to address such challenges. Our analysis shows that for a given number of quantization bits, each block of Transformer contributes to translation quality and inference computations in different manners. Moreover, even inside an embedding block, each word presents vastly different contributions. Correspondingly, we…
How to efficiently serve ever-larger trained natural language models in practice has become exceptionally challenging even for powerful cloud servers due to their prohibitive memory/computation requirements. In this work, we present an efficient and affordable post-training quantization approach to compress large Transformer-based models, termed as ZeroQuant. ZeroQuant is an end-to-end quantization and inference pipeline with three main components: (1) a fine-grained hardware-friendly quantization scheme for both weight and activations; (2) a novel affordable layer-by-layer knowledge distillat…
University of Warsaw（华沙大学）；Google Research（谷歌研究院）；OpenAI
Large Transformer models yield impressive results on many tasks, but are expensive to train, or even fine-tune, and so slow at decoding that their use and study becomes out of reach. We address this problem by leveraging sparsity. We study sparse variants for all layers in the Transformer and propose Scaling Transformers, a family of next generation Transformer models that use sparse layers to scale efficiently and perform unbatched decoding much faster than the standard Transformer as we scale up the model size. Surprisingly, the sparse layers are enough to obtain the same perplexity as the s…
University of Cambridge（剑桥大学）；Microsoft（微软）；Imperial College London（帝国理工学院）
Large Language Models (LLMs) have demonstrated impressive performance on a range of Natural Language Processing (NLP) tasks. Unfortunately, the immense amount of computations and memory accesses required for LLM training makes them prohibitively expensive in terms of hardware cost, and thus challenging to deploy in use cases such as on-device learning. In this paper, motivated by the observation that LLM training is memory-bound, we propose a novel dynamic quantization strategy, termed Dynamic Stashing Quantization (DSQ), that puts a special focus on reducing the memory operations, but also en…
TWEO: Transformers Without Extreme Outliers Enables FP8 Training And Quantization For Dummies
Nationa Key Laboratory for Novel Software Technology（新型软件技术国家实验室）；School of Artificial Intelligence（人工智能学院）；Zhongguancun Academy（中关村学院）；University of Science and Technology of China（中国科学技术大学）
