SparQ Attention: Bandwidth-Efficient LLM Inference

Generative large language models (LLMs) have opened up numerous novel possibilities, but due to their significant computational requirements their ubiquitous use remains challenging. Some of the most useful applications require processing large numbers of samples at a time and using long contexts, both significantly increasing the memory communication load of the models. We introduce SparQ Attention, a technique for increasing the inference throughput of LLMs by reducing the memory bandwidth requirements within the attention blocks through selective fetching of the cached history. Our proposed technique can be applied directly to off-the-shelf LLMs during inference, without requiring any modification to the pre-training setup or additional fine-tuning. We show how SparQ Attention can decrease the attention memory bandwidth requirements up to eight times without any loss in accuracy by evaluating Llama 2 and Pythia models on a wide range of downstream tasks.

PopSparse: Accelerated block sparse matrix multiplication on IPU

Reducing the computational cost of running large scale neural networks using sparsity has attracted great attention in the deep learning community. While much success has been achieved in reducing FLOP and parameter counts while maintaining acceptable task performance, achieving actual speed improvements has typically been much more difficult, particularly on general purpose accelerators (GPAs) such as NVIDIA GPUs using low precision number formats. In this work we introduce PopSparse, a library that enables fast sparse operations on Graphcore IPUs by leveraging both the unique hardware characteristics of IPUs as well as any block structure defined in the data. We target two different types of sparsity: static, where the sparsity pattern is fixed at compile-time; and dynamic, where it can change each time the model is run. We present benchmark results for matrix multiplication for both of these modes on IPU with a range of block sizes, matrix sizes and densities. Results indicate that the PopSparse implementations are faster than dense matrix multiplications on IPU at a range of sparsity levels with large matrix size and block size. Furthermore, static sparsity in general outperforms dynamic sparsity. While previous work on GPAs has shown speedups only for very high sparsity (typically 99\% and above), the present work demonstrates that our static sparse implementation outperforms equivalent dense calculations in FP16 at lower sparsity (around 90%). IPU code is available to view and run at ipu.dev/sparsity-benchmarks, GPU code will be made available shortly.

Unit Scaling: Out-of-the-Box Low-Precision Training

We present unit scaling, a paradigm for designing deep learning models that simplifies the use of low-precision number formats. Training in FP16 or the recently proposed FP8 formats offers substantial efficiency gains, but can lack sufficient range for out-of-the-box training. Unit scaling addresses this by introducing a principled approach to model numerics: seeking unit variance of all weights, activations and gradients at initialisation. Unlike alternative methods, this approach neither requires multiple training runs to find a suitable scale nor has significant computational overhead. We demonstrate the efficacy of unit scaling across a range of models and optimisers. We further show that existing models can be adapted to be unit-scaled, training BERT-Large in FP16 and then FP8 with no degradation in accuracy.

BESS: Balanced Entity Sampling and Sharing for Large-Scale Knowledge Graph Completion

We present the award-winning submission to the WikiKG90Mv2 track of OGB-LSC@NeurIPS 2022. The task is link-prediction on the large-scale knowledge graph WikiKG90Mv2, consisting of 90M+ nodes and 600M+ edges. Our solution uses a diverse ensemble of $85$ Knowledge Graph Embedding models combining five different scoring functions (TransE, TransH, RotatE, DistMult, ComplEx) and two different loss functions (log-sigmoid, sampled softmax cross-entropy). Each individual model is trained in parallel on a Graphcore Bow Pod$_{16}$ using BESS (Balanced Entity Sampling and Sharing), a new distribution framework for KGE training and inference based on balanced collective communications between workers. Our final model achieves a validation MRR of 0.2922 and a test-challenge MRR of 0.2562, winning the first place in the competition. The code is publicly available at: https://github.com/graphcore/distributed-kge-poplar/tree/2022-ogb-submission.

Towards Structured Dynamic Sparse Pre-Training of BERT

Identifying algorithms for computational efficient unsupervised training of large language models is an important and active area of research. In this work, we develop and study a straightforward, dynamic always-sparse pre-training approach for BERT language modeling task, which leverages periodic compression steps based on magnitude pruning followed by random parameter re-allocation. This approach enables us to achieve Pareto improvements in terms of the number of floating-point operations (FLOPs) over statically sparse and dense models across a broad spectrum of network sizes. Furthermore, we demonstrate that training remains FLOP-efficient when using coarse-grained block sparsity, making it particularly promising for efficient execution on modern hardware accelerators.

GroupBERT: Enhanced Transformer Architecture with Efficient Grouped Structures

Attention based language models have become a critical component in state-of-the-art natural language processing systems. However, these models have significant computational requirements, due to long training times, dense operations and large parameter count. In this work we demonstrate a set of modifications to the structure of a Transformer layer, producing a more efficient architecture. First, we add a convolutional module to complement the self-attention module, decoupling the learning of local and global interactions. Secondly, we rely on grouped transformations to reduce the computational cost of dense feed-forward layers and convolutions, while preserving the expressivity of the model. We apply the resulting architecture to language representation learning and demonstrate its superior performance compared to BERT models of different scales. We further highlight its improved efficiency, both in terms of floating-point operations (FLOPs) and time-to-train.

N-grams Bayesian Differential Privacy

Differential privacy has gained popularity in machine learning as a strong privacy guarantee, in contrast to privacy mitigation techniques such as k-anonymity. However, applying differential privacy to n-gram counts significantly degrades the utility of derived language models due to their large vocabularies. We propose a differential privacy mechanism that uses public data as a prior in a Bayesian setup to provide tighter bounds on the privacy loss metric epsilon, and thus better privacy-utility trade-offs. It first transforms the counts to log space, approximating the distribution of the public and private data as Gaussian. The posterior distribution is then evaluated and softmax is applied to produce a probability distribution. This technique achieves up to 85% reduction in KL divergence compared to previously known mechanisms at epsilon equals 0.1. We compare our mechanism to k-anonymity in a n-gram language modelling task and show that it offers competitive performance at large vocabulary sizes, while also providing superior privacy protection.