Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userHardware Acceleration of Fully Quantized BERT for Efficient Natural Language Processing
200
0
0.0
(
0
)
Added by
Zejian Liu
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
BERT is the most recent Transformer-based model that achieves state-of-the-art performance in various NLP tasks. In this paper, we investigate the hardware acceleration of BERT on FPGA for edge computing. To tackle the issue of huge computational complexity and memory footprint, we propose to fully quantize the BERT (FQ-BERT), including weights, activations, softmax, layer normalization, and all the intermediate results. Experiments demonstrate that the FQ-BERT can achieve 7.94x compression for weights with negligible performance loss. We then propose an accelerator tailored for the FQ-BERT and evaluate on Xilinx ZCU102 and ZCU111 FPGA. It can achieve a performance-per-watt of 3.18 fps/W, which is 28.91x and 12.72x over Intel(R) Core(TM) i7-8700 CPU and NVIDIA K80 GPU, respectively.
rate research
Read More
Transformers are ubiquitous in Natural Language Processing (NLP) tasks, but they are difficult to be deployed on hardware due to the intensive computation. To enable low-latency inference on resource-constrained hardware platforms, we propose to design Hardware-Aware Transformers (HAT) with neural architecture search. We first construct a large design space with $textit{arbitrary encoder-decoder attention}$ and $textit{heterogeneous layers}$. Then we train a $textit{SuperTransformer}$ that covers all candidates in the design space, and efficiently produces many $textit{SubTransformers}$ with weight sharing. Finally, we perform an evolutionary search with a hardware latency constraint to find a specialized $textit{SubTransformer}$ dedicated to run fast on the target hardware. Extensive experiments on four machine translation tasks demonstrate that HAT can discover efficient models for different hardware (CPU, GPU, IoT device). When running WMT14 translation task on Raspberry Pi-4, HAT can achieve $textbf{3}times$ speedup, $textbf{3.7}times$ smaller size over baseline Transformer; $textbf{2.7}times$ speedup, $textbf{3.6}times$ smaller size over Evolved Transformer with $textbf{12,041}times$ less search cost and no performance loss. HAT code is https://github.com/mit-han-lab/hardware-aware-transformers.git
Convolutional neural network (CNN) inference on mobile devices demands efficient hardware acceleration of low-precision (INT8) general matrix multiplication (GEMM). Exploiting data sparsity is a common approach to further accelerate GEMM for CNN inference, and in particular, structural sparsity has the advantages of predictable load balancing and very low index overhead. In this paper, we address a key architectural challenge with structural sparsity: how to provide support for a range of sparsity levels while maintaining high utilization of the hardware. We describe a time unrolled formulation of variable density-bound block (VDBB) sparsity that allows for a configurable number of non-zero elements per block, at constant utilization. We then describe a systolic array microarchitecture that implements this scheme, with two data reuse optimizations. Firstly, we increase reuse in both operands and partial products by increasing the number of MACs per PE. Secondly, we introduce a novel approach of moving the IM2COL transform into the hardware, which allows us to achieve a 3x data bandwidth expansion just before the operands are consumed by the datapath, reducing the SRAM power consumption. The optimizations for weight sparsity, activation sparsity and data reuse are all interrelated and therefore the optimal combination is not obvious. Therefore, we perform an design space evaluation to find the pareto-optimal design characteristics. The resulting design achieves 16.8 TOPS/W in 16nm with modest 50% model sparsity and scales with model sparsity up to 55.7TOPS/W at 87.5%. As well as successfully demonstrating the variable DBB technique, this result significantly outperforms previously reported sparse CNN accelerators.
Recently, large pre-trained neural language models have attained remarkable performance on many downstream natural language processing (NLP) applications via fine-tuning. In this paper, we target at how to further improve the token representations on the language models. We, therefore, propose a simple yet effective plug-and-play module, Sequential Attention Module (SAM), on the token embeddings learned from a pre-trained language model. Our proposed SAM consists of two main attention modules deployed sequentially: Feature-wise Attention Module (FAM) and Token-wise Attention Module (TAM). More specifically, FAM can effectively identify the importance of features at each dimension and promote the effect via dot-product on the original token embeddings for downstream NLP applications. Meanwhile, TAM can further re-weight the features at the token-wise level. Moreover, we propose an adaptive filter on FAM to prevent noise impact and increase information absorption. Finally, we conduct extensive experiments to demonstrate the advantages and properties of our proposed SAM. We first show how SAM plays a primary role in the champion solution of two subtasks of SemEval21 Task 7. After that, we apply SAM on sentiment analysis and three popular NLP tasks and demonstrate that SAM consistently outperforms the state-of-the-art baselines.
Many search systems work with large amounts of natural language data, e.g., search queries, user profiles, and documents. Building a successful search system requires a thorough understanding of textual data semantics, where deep learning based natural language processing techniques (deep NLP) can be of great help. In this paper, we introduce a comprehensive study for applying deep NLP techniques to five representative tasks in search systems: query intent prediction (classification), query tagging (sequential tagging), document ranking (ranking), query auto completion (language modeling), and query suggestion (sequence to sequence). We also introduce BERT pre-training as a sixth task that can be applied to many of the other tasks. Through the model design and experiments of the six tasks, readers can find answers to four important questions: (1). When is deep NLP helpful/not helpful in search systems? (2). How to address latency challenges? (3). How to ensure model robustness? This work builds on existing efforts of LinkedIn search, and is tested at scale on LinkedIns commercial search engines. We believe our experiences can provide useful insights for the industry and research communities.
The TSNLP project has investigated various aspects of the construction, maintenance and application of systematic test suites as diagnostic and evaluation tools for NLP applications. The paper summarizes the motivation and main results of the project: besides the solid methodological foundation, TSNLP has produced substantial multi-purpose and multi-user test suites for three European languages together with a set of specialized tools that facilitate the construction, extension, maintenance, retrieval, and customization of the test data. As TSNLP results, including the data and technology, are made publicly available, the project presents a valuable linguistic resourc e that has the potential of providing a wide-spread pre-standard diagnostic and evaluation tool for both developers and users of NLP applications.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
