On-Chip Learning via Transformer In-Context Learning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Finkbeiner, Jan, Neftci, Emre
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913541925109760
author Finkbeiner, Jan
Neftci, Emre
author_facet Finkbeiner, Jan
Neftci, Emre
contents Autoregressive decoder-only transformers have become key components for scalable sequence processing and generation models. However, the transformer's self-attention mechanism requires transferring prior token projections from the main memory at each time step (token), thus severely limiting their performance on conventional processors. Self-attention can be viewed as a dynamic feed-forward layer, whose matrix is input sequence-dependent similarly to the result of local synaptic plasticity. Using this insight, we present a neuromorphic decoder-only transformer model that utilizes an on-chip plasticity processor to compute self-attention. Interestingly, the training of transformers enables them to ``learn'' the input context during inference. We demonstrate this in-context learning ability of transformers on the Loihi 2 processor by solving a few-shot classification problem. With this we emphasize the importance of pretrained models especially their ability to find simple, local, backpropagation free, learning rules enabling on-chip learning and adaptation in a hardware friendly manner.
format Preprint
id arxiv_https___arxiv_org_abs_2410_08711
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On-Chip Learning via Transformer In-Context Learning
Finkbeiner, Jan
Neftci, Emre
Neural and Evolutionary Computing
Autoregressive decoder-only transformers have become key components for scalable sequence processing and generation models. However, the transformer's self-attention mechanism requires transferring prior token projections from the main memory at each time step (token), thus severely limiting their performance on conventional processors. Self-attention can be viewed as a dynamic feed-forward layer, whose matrix is input sequence-dependent similarly to the result of local synaptic plasticity. Using this insight, we present a neuromorphic decoder-only transformer model that utilizes an on-chip plasticity processor to compute self-attention. Interestingly, the training of transformers enables them to ``learn'' the input context during inference. We demonstrate this in-context learning ability of transformers on the Loihi 2 processor by solving a few-shot classification problem. With this we emphasize the importance of pretrained models especially their ability to find simple, local, backpropagation free, learning rules enabling on-chip learning and adaptation in a hardware friendly manner.
title On-Chip Learning via Transformer In-Context Learning
topic Neural and Evolutionary Computing
url https://arxiv.org/abs/2410.08711