Parallelizing Linear Transformers with the Delta Rule over Sequence Length

• URL: http://arxiv.org/abs/2406.06484v4
• Date: Tue, 05 Nov 2024 16:48:53 GMT
• Title: Parallelizing Linear Transformers with the Delta Rule over Sequence Length
• Authors: Songlin Yang, Bailin Wang, Yu Zhang, Yikang Shen, Yoon Kim,
• Abstract summary: This work describes a hardware-efficient algorithm for training linear transformers with the delta rule. We train a 1.3B model for 100B tokens and find that it outperforms recent linear-time baselines.
• Score: 49.88826673324244
• License:
• Abstract: Transformers with linear attention (i.e., linear transformers) and state-space models have recently been suggested as a viable linear-time alternative to transformers with softmax attention. However, these models still underperform transformers especially on tasks that require in-context retrieval. While more expressive variants of linear transformers which replace the additive update in linear transformers with the delta rule (DeltaNet) have been found to be more effective at associative recall, existing algorithms for training such models do not parallelize over sequence length and are thus inefficient to train on modern hardware. This work describes a hardware-efficient algorithm for training linear transformers with the delta rule, which exploits a memory-efficient representation for computing products of Householder matrices. This algorithm allows us to scale up DeltaNet to standard language modeling settings. We train a 1.3B model for 100B tokens and find that it outperforms recent linear-time baselines such as Mamba and GLA in terms of perplexity and zero-shot performance on downstream tasks. We also experiment with two hybrid models which combine DeltaNet layers with (1) sliding-window attention layers every other layer or (2) two global attention layers, and find that these hybrids outperform strong transformer baselines.

Related papers

• The Mamba in the Llama: Distilling and Accelerating Hybrid Models [76.64055251296548]
  We show that it is feasible to distill large Transformers into linear RNNs by reusing the linear projection weights from attention layers with academic GPU resources. The resulting hybrid model, which incorporates a quarter of the attention layers, achieves performance comparable to the original Transformer in chat benchmarks.
  arXiv  Detail & Related papers  (2024-08-27T17:56:11Z)
• Your Transformer is Secretly Linear [7.935853865895353]
  We analyze embedding transformations between sequential layers, uncovering a near-perfect linear relationship. We show that removing or linearly approximating some of the most linear blocks of transformers does not affect significantly the loss or model performance. In our pretraining experiments on smaller models we introduce a cosine-similarity-based regularization, aimed at reducing layer linearity.
  arXiv  Detail & Related papers  (2024-05-19T22:44:00Z)
• Gated Linear Attention Transformers with Hardware-Efficient Training [60.670102007737476]
  This work describes a hardware-efficient algorithm for linear attention that trades off memory movement against parallelizability. We then generalize this algorithm to a more expressive variant of linear attention with data-dependent gates. When used as a replacement for the standard attention layer in Transformers, the resulting gated linear attention Transformer is found to perform competitively.
  arXiv  Detail & Related papers  (2023-12-11T18:51:59Z)
• Emergent Agentic Transformer from Chain of Hindsight Experience [96.56164427726203]
  We show that a simple transformer-based model performs competitively with both temporal-difference and imitation-learning-based approaches. This is the first time that a simple transformer-based model performs competitively with both temporal-difference and imitation-learning-based approaches.
  arXiv  Detail & Related papers  (2023-05-26T00:43:02Z)
• Momentum Transformer: Closing the Performance Gap Between Self-attention and Its Linearization [31.28396970291575]
  Leveraging techniques include sparse and linear attention and hashing tricks; efficient transformers have been proposed to reduce the quadratic complexity of transformers but significantly degrade the accuracy. We first interpret the linear attention and residual connections in computing the attention map as gradient descent steps. We then introduce momentum into these components and propose the emphmomentum transformer, which utilizes momentum to improve the accuracy of linear transformers while maintaining linear memory and computational complexities.
  arXiv  Detail & Related papers  (2022-08-01T02:37:49Z)
• Finetuning Pretrained Transformers into RNNs [81.72974646901136]
  Transformers have outperformed recurrent neural networks (RNNs) in natural language generation. A linear-complexity recurrent variant has proven well suited for autoregressive generation. This work aims to convert a pretrained transformer into its efficient recurrent counterpart.
  arXiv  Detail & Related papers  (2021-03-24T10:50:43Z)
• Long Range Arena: A Benchmark for Efficient Transformers [115.1654897514089]
  Long-rangearena benchmark is a suite of tasks consisting of sequences ranging from $1K$ to $16K$ tokens. We systematically evaluate ten well-established long-range Transformer models on our newly proposed benchmark suite.
  arXiv  Detail & Related papers  (2020-11-08T15:53:56Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.