Scan and Snap: Understanding Training Dynamics and Token Composition in 1-layer Transformer

• URL: http://arxiv.org/abs/2305.16380v4
• Date: Mon, 30 Oct 2023 17:32:08 GMT
• Title: Scan and Snap: Understanding Training Dynamics and Token Composition in 1-layer Transformer
• Authors: Yuandong Tian, Yiping Wang, Beidi Chen, Simon Du
• Abstract summary: Transformer architecture has shown impressive performance in multiple research domains. We analyze its SGD training dynamics for the task of next token prediction. We prove that self-attention acts as a emphdiscriminative scanning algorithm.
• Score: 37.37547759817417
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Transformer architecture has shown impressive performance in multiple research domains and has become the backbone of many neural network models. However, there is limited understanding on how it works. In particular, with a simple predictive loss, how the representation emerges from the gradient \emph{training dynamics} remains a mystery. In this paper, for 1-layer transformer with one self-attention layer plus one decoder layer, we analyze its SGD training dynamics for the task of next token prediction in a mathematically rigorous manner. We open the black box of the dynamic process of how the self-attention layer combines input tokens, and reveal the nature of underlying inductive bias. More specifically, with the assumption (a) no positional encoding, (b) long input sequence, and (c) the decoder layer learns faster than the self-attention layer, we prove that self-attention acts as a \emph{discriminative scanning algorithm}: starting from uniform attention, it gradually attends more to distinct key tokens for a specific next token to be predicted, and pays less attention to common key tokens that occur across different next tokens. Among distinct tokens, it progressively drops attention weights, following the order of low to high co-occurrence between the key and the query token in the training set. Interestingly, this procedure does not lead to winner-takes-all, but decelerates due to a \emph{phase transition} that is controllable by the learning rates of the two layers, leaving (almost) fixed token combination. We verify this \textbf{\emph{scan and snap}} dynamics on synthetic and real-world data (WikiText).

Related papers

• FIRP: Faster LLM inference via future intermediate representation prediction [54.897493351694195]
  FIRP generates multiple tokens instead of one at each decoding step. We conduct extensive experiments, showing a speedup ratio of 1.9x-3x in several models and datasets.
  arXiv  Detail & Related papers  (2024-10-27T15:53:49Z)
• Looking Beyond The Top-1: Transformers Determine Top Tokens In Order [13.032106683136394]
  We analyze the computation performed by Transformers in the layers after the top-1 prediction has become fixed. We find that these saturation events happen in order of the corresponding tokens' ranking. We propose an underlying mechanism of task transition for this sequential saturation.
  arXiv  Detail & Related papers  (2024-10-26T16:00:38Z)
• ToSA: Token Selective Attention for Efficient Vision Transformers [50.13756218204456]
  ToSA is a token selective attention approach that can identify tokens that need to be attended as well as those that can skip a transformer layer. We show that ToSA can significantly reduce computation costs while maintaining accuracy on the ImageNet classification benchmark.
  arXiv  Detail & Related papers  (2024-06-13T05:17:21Z)
• Token-Label Alignment for Vision Transformers [93.58540411138164]
  Data mixing strategies (e.g., CutMix) have shown the ability to greatly improve the performance of convolutional neural networks (CNNs) We identify a token fluctuation phenomenon that has suppressed the potential of data mixing strategies. We propose a token-label alignment (TL-Align) method to trace the correspondence between transformed tokens and the original tokens to maintain a label for each token.
  arXiv  Detail & Related papers  (2022-10-12T17:54:32Z)
• DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification [134.9393799043401]
  We propose a dynamic token sparsification framework to prune redundant tokens based on the input. By hierarchically pruning 66% of the input tokens, our method greatly reduces 31%37% FLOPs and improves the throughput by over 40%. DynamicViT models can achieve very competitive complexity/accuracy trade-offs compared to state-of-the-art CNNs and vision transformers on ImageNet.
  arXiv  Detail & Related papers  (2021-06-03T17:57:41Z)
• KVT: k-NN Attention for Boosting Vision Transformers [44.189475770152185]
  We propose a sparse attention scheme, dubbed k-NN attention, for boosting vision transformers. The proposed k-NN attention naturally inherits the local bias of CNNs without introducing convolutional operations. We verify, both theoretically and empirically, that $k$-NN attention is powerful in distilling noise from input tokens and in speeding up training.
  arXiv  Detail & Related papers  (2021-05-28T06:49:10Z)
• Fast End-to-End Speech Recognition via a Non-Autoregressive Model and Cross-Modal Knowledge Transferring from BERT [72.93855288283059]
  We propose a non-autoregressive speech recognition model called LASO (Listen Attentively, and Spell Once) The model consists of an encoder, a decoder, and a position dependent summarizer (PDS)
  arXiv  Detail & Related papers  (2021-02-15T15:18:59Z)

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.