Guardado en:
Detalles Bibliográficos
Autores principales: Xiao, Emily, Li, Chin-Jou, Zhang, Yilin, Neubig, Graham, Bertsch, Amanda
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2503.08640
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909541318262784
author Xiao, Emily
Li, Chin-Jou
Zhang, Yilin
Neubig, Graham
Bertsch, Amanda
author_facet Xiao, Emily
Li, Chin-Jou
Zhang, Yilin
Neubig, Graham
Bertsch, Amanda
contents Many-shot in-context learning has recently shown promise as an alternative to finetuning, with the major advantage that the same model can be served for multiple tasks. However, this shifts the computational burden from training-time to inference-time, making deployment of many-shot ICL challenging to justify in-practice. This cost is further increased if a custom demonstration set is retrieved for each inference example. We present Dynamic Block-Sparse Attention, a training-free framework for retrieval-based many-shot in-context learning. By combining carefully designed block-sparse attention and retrieval of cached groups of demonstrations, we achieve comparable per-example latency to finetuning while maintaining on average >95% of the best method's accuracy across strong ICL and finetuning baselines. We hope that this will further enable the deployment of many-shot ICL at scale.
format Preprint
id arxiv_https___arxiv_org_abs_2503_08640
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Efficient Many-Shot In-Context Learning with Dynamic Block-Sparse Attention
Xiao, Emily
Li, Chin-Jou
Zhang, Yilin
Neubig, Graham
Bertsch, Amanda
Computation and Language
Many-shot in-context learning has recently shown promise as an alternative to finetuning, with the major advantage that the same model can be served for multiple tasks. However, this shifts the computational burden from training-time to inference-time, making deployment of many-shot ICL challenging to justify in-practice. This cost is further increased if a custom demonstration set is retrieved for each inference example. We present Dynamic Block-Sparse Attention, a training-free framework for retrieval-based many-shot in-context learning. By combining carefully designed block-sparse attention and retrieval of cached groups of demonstrations, we achieve comparable per-example latency to finetuning while maintaining on average >95% of the best method's accuracy across strong ICL and finetuning baselines. We hope that this will further enable the deployment of many-shot ICL at scale.
title Efficient Many-Shot In-Context Learning with Dynamic Block-Sparse Attention
topic Computation and Language
url https://arxiv.org/abs/2503.08640