We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Bigger&Faster: Two-stage Neural Architecture Search for Quantized Transformer Models

Abstract: Neural architecture search (NAS) for transformers has been used to create state-of-the-art models that target certain latency constraints. In this work we present Bigger&Faster, a novel quantization-aware parameter sharing NAS that finds architectures for 8-bit integer (int8) quantized transformers. Our results show that our method is able to produce BERT models that outperform the current state-of-the-art technique, AutoTinyBERT, at all latency targets we tested, achieving up to a 2.68% accuracy gain. Additionally, although the models found by our technique have a larger number of parameters than their float32 counterparts, due to their parameters being int8, they have significantly smaller memory footprints.
Subjects: Machine Learning (cs.LG)
Cite as: arXiv:2209.12127 [cs.LG]
  (or arXiv:2209.12127v1 [cs.LG] for this version)

Submission history

From: Yuji Chai [view email]
[v1] Sun, 25 Sep 2022 02:56:01 GMT (1135kb,D)

Link back to: arXiv, form interface, contact.