David Grangier
Machine learning researcher at Apple; co-author of the fairseq sequence modeling toolkit
David Grangier is a researcher with the Machine Learning Group at Apple, with interests in large-scale machine learning for text generation, information retrieval, and speech recognition. Grangier's vitae lists earlier research roles at Google Brain from 2018, Facebook AI Research from 2014 to 2018, and Microsoft Research from 2012 to 2014, along with work on fairseq, a sequence-to-sequence toolkit built with Myle Ott, Sergey Edunov, Michael Auli, and others. Recent Apple work listed on Grangier's homepage includes an ICLR 2026 paper on pretraining small language models with hierarchical parametric memory banks, motivated by the limits of storing world knowledge in the parameters of models for edge devices.1235
Work
fairseq (external site: github.com)Open-source code
Co-author of the 2019 fairseq toolkit paper, listed with Google Brain, and named in Grangier's vitae among the people who built fairseq from 2016 to 2018; the repository is released under the MIT license.324
Pretraining with hierarchical memories: separating long-tail and common knowledge (external site: arxiv.org)Published research
Co-author of the ICLR 2026 paper introducing small language models that draw on large hierarchical parametric memory banks, aimed at models for edge devices.51
Other documented roles
Roles the sources document but that are not confirmed as current.
- Machine learning research (from 2018), Google Brain23
- Machine learning research (2014–2018), Facebook AI Research2
- Machine learning research (2012–2014), Microsoft Research2
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.