Shawn Tan

Scaling Stick-Breaking Attention: An Efficient Implementation and In-depth Study

Songlin Yang

Rameswar Panda

The self-attention mechanism traditionally relies on the softmax operator, necessitating positional embeddings like RoPE, or position biases… (voir plus) to account for token order. But current methods using still face length generalisation challenges. We investigate an alternative attention mechanism based on the stick-breaking process in larger scale settings. The method works as follows: For each token before the current, we determine a break point, which represents the proportion of the stick, the weight of the attention, to allocate to the current token. We repeat this on the remaining stick, until all tokens are allocated a weight, resulting in a sequence of attention weights. This process naturally incorporates recency bias, which has linguistic motivations for grammar parsing (Shen et al., 2017). We study the implications of replacing the conventional softmax-based attention mechanism with stick-breaking attention. We then discuss implementation of numerically stable stick-breaking attention and adapt Flash Attention to accommodate this mechanism. When used as a drop-in replacement for current softmax+RoPE attention systems, we find that stick-breaking attention performs competitively with current methods on length generalisation and downstream tasks. Stick-breaking also performs well at length generalisation, allowing a model trained with

2025-01-22

ICLR.cc/2025/Conference (poster)

Stick-breaking Attention

Songlin Yang

Rameswar Panda

2024-10-23

ArXiv (prépublication)

Stick-breaking Attention

Songlin Yang

Rameswar Panda

2024-10-23

ArXiv (prépublication)

Stick-breaking Attention

Songlin Yang

Rameswar Panda

2024-10-23

ArXiv (prépublication)

Stick-breaking Attention

Songlin Yang

Rameswar Panda

2024-10-23

ArXiv (prépublication)

Stick-breaking Attention

Songlin Yang

Rameswar Panda

2024-10-23

ArXiv (prépublication)

Scattered Mixture-of-Experts Implementation

Rameswar Panda

ScatterMoE is an implementation of Sparse Mixture-of-Experts (SMoE) on GPUs. ScatterMoE builds upon techniques in existing implementations, … (voir plus)and overcoming some of the current limitations to improve batched inference, training speed, and memory footprint. This implementation achieves this by avoiding padding and making excessive copies of the input. We also fuse expert linear transforms and reordering operations with ParallelLinear, a module that can be used to extend the concept of SMoEs. We benchmark our implementation against Megablocks, and show that it enables a higher throughput and lower memory footprint. We also show how ParallelLinear enables extension of the Mixture-of-Experts concept by demonstrating with an implementation of Mixture-of-Attention.

2024-07-10

colmweb.org/COLM/2024/Conference (accepté)

Scattered Mixture-of-Experts Implementation

Rameswar Panda

We present ScatterMoE, an implementation of Sparse Mixture-of-Experts (SMoE) on GPUs. ScatterMoE builds upon existing implementations, and o… (voir plus)vercoming some of the limitations to improve inference and training speed, and memory footprint. This implementation achieves this by avoiding padding and making excessive copies of the input. We introduce ParallelLinear, the main component we use to build our implementation and the various kernels used to speed up the operation. We benchmark our implementation against Megablocks, and show that it enables a higher throughput and lower memory footprint. We also show how ParallelLinear enables extension of the Mixture-of-Experts concept by demonstrating with an implementation of Mixture of Attention.

2024-03-13

ArXiv (prépublication)

Sparse Universal Transformer

Zhenfang Chen

Chuang Gan

The Universal Transformer (UT) is a variant of the Transformer that shares parameters across its layers and is Turing-complete under certain… (voir plus) assumptions. Empirical evidence also shows that UTs have better compositional generalization than Vanilla Transformers (VTs) in formal language tasks. The parameter-sharing also affords it better parameter efficiency than VTs. Despite its many advantages, most state-of-the-art NLP systems use VTs as their backbone model instead of UTs. This is mainly because scaling UT parameters is more compute and memory intensive than scaling up a VT. This paper proposes the Sparse Universal Transformer (SUT), which leverages Sparse Mixture of Experts (SMoE) to reduce UT's computation complexity while retaining its parameter efficiency and generalization ability. Experiments show that SUT combines the best of both worlds, achieving strong generalization results on formal language tasks (Logical inference and CFQ) and impressive parameter and computation efficiency on standard natural language benchmarks like WMT'14.

2023-10-07

EMNLP/2023/Conference (accepté)

Unsupervised Dependency Graph Network

Alessandro Sordoni

Peng Li

Jie Zhou

2022-05-01

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (publié)

Learning to Dequantise with Truncated Flows

Dequantisation is a general technique used for transforming data described by a discrete random variable x into a continuous (latent) random… (voir plus) variable z, for the purpose of it being modeled by likelihood-based density models. Dequantisation was first introduced in the context of ordinal data, such as image pixel values. However, when the data is categorical, the dequantisation scheme is not obvious. We learn such a dequantisation scheme q(z|x), using variational inference with TRUncated FLows (TRUFL) — a novel flow-based model that allows the dequantiser to have a learnable truncated support. Unlike previous work, the TRUFL dequantiser is (i) capable of embedding the data losslessly in certain cases, since the truncation allows the conditional distributions q(z|x) to have non-overlapping bounded supports, while being (ii) trainable with back-propagation. Addtionally, since the support of the marginal q(z) is bounded and the support of prior p(z) is not, we propose to renormalise the prior distribution over the support of q(z). We derive a lower bound for training, and propose a rejection sampling scheme to account for the invalid samples. Experimentally, we benchmark TRUFL on constrained generation tasks, and find that it outperforms prior approaches. In addition, we find that rejection sampling results in higher validity for the constrained problems.

2022-01-01

International Conference on Learning Representations (publié)