Builders & open source
Ruoming Pang
Listed with Meta Platforms on the ScaleAcross Explorer paper (May 2026); GitHub profile lists Meta, Meta
- Listed with Meta Platforms on the ScaleAcross Explorer paper (May 2026); GitHub profile lists Meta, Meta12
Ruoming Pang's GitHub profile lists Meta as Pang's organization, and Pang is listed with Meta Platforms on a May 2026 paper about communication optimization for large-scale model training across data centers. At Apple, Pang was the overall lead of the foundation models contributors on the 2024 Apple Intelligence Foundation Language Models report, which describes an on-device model of about 3 billion parameters and a larger server model, and is the corresponding author of the paper on AXLearn, Apple's open-source system for modular, hardware-agnostic training of large models. Earlier, Pang was listed with Google Inc. as an author of the 2020 Conformer paper on convolution-augmented Transformers for speech recognition.12345
At a glance
- Current role
- Listed with Meta Platforms on the ScaleAcross Explorer paper (May 2026); GitHub profile lists Meta
- Company
- Meta
- Sector
- Company
- Education
- Princeton
- Work listed
- 3
- Sources cited
- 9
- Profile reviewed
Research at Google
Ruoming Pang's earlier papers, written at Google, centre on speech recognition and sequence-model infrastructure. Pang is among the authors of the 2019 paper describing Lingvo, a modular and scalable TensorFlow framework for sequence-to-sequence modeling, and co-authored the 2020 ContextNet and Conformer papers, which improve speech recognition with global context in convolutional networks and by combining convolution with Transformer self-attention.675
Foundation models and training systems
At Apple, Pang was the overall lead of the foundation models contributors on the July 2024 Apple Intelligence Foundation Language Models report, which describes an on-device model of about 3 billion parameters and a larger server model. Pang is also the corresponding author of the 2025 paper on AXLearn, Apple's open-source library for modular, hardware-agnostic training of large models.348
Pang's recent papers address the systems cost of large models. Parallel Track Transformers (2026) reduces synchronization in multi-GPU inference, and ScaleAcross Explorer (May 2026), on which Pang is listed with Meta Platforms, studies communication optimization for training spread across data center buildings and regions.921
Work and contributions
AXLearn Open-source code
Corresponding author of the AXLearn paper, which describes Apple's production system for scalable, modular, hardware-agnostic training of large deep learning models; the apple/axlearn repository is released under the Apache 2.0 license.48
Apple Intelligence Foundation Language Models Proprietary product
Listed as overall lead among the foundation models contributors of Apple's 2024 report on the language models behind Apple Intelligence.3
Conformer: Convolution-augmented Transformer for Speech Recognition Published research
Co-author of the 2020 Conformer paper, listed with Google Inc.5
Timeline
- Co-authored the Lingvo framework paper on sequence-to-sequence modeling.6
- Co-authored the ContextNet and Conformer speech recognition papers.75
- Led the foundation models contributors on the Apple Intelligence Foundation Language Models report.3
- Was corresponding author of the AXLearn paper on modular large-model training.4
- Co-authored Parallel Track Transformers on fast GPU inference with reduced synchronization.9
- Co-authored the ScaleAcross Explorer paper, listed with Meta Platforms.2
Other documented roles
Roles the sources document that are not confirmed as current.
Where they work
- Meta 106 people in the wiki
Related people
- Emmanuel DupouxAlso at Meta
- Yu-An ChungAlso at Meta
- David GrangierAlso at Meta
- Edouard GraveAlso at Meta
- Myle OttAlso at Meta
- Yonghui WuLanguage models · Training systems
- Lucas AtkinsLanguage models · Training systems
- Koray KavukcuogluLanguage models · Speech and audio
- Torsten ScholakLanguage models · Training systems
- Rohit PrasadLanguage models · Speech and audio
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
9 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- ruomingp (Ruoming Pang) on GitHub
- ScaleAcross Explorer: Exploring Communication Optimization for Scale-Across AI Model Training (arXiv 2605.24326)
- Apple Intelligence Foundation Language Models (arXiv 2407.21075)
- AXLearn: Modular, Hardware-Agnostic Large Model Training (arXiv 2507.05411)
- Conformer: Convolution-augmented Transformer for Speech Recognition (arXiv 2005.08100)
- Lingvo: a Modular and Scalable Framework for Sequence-to-Sequence Modeling (arXiv 1902.08295)
- ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context (arXiv 2005.03191)
- apple/axlearn LICENSE (Apache 2.0)
- Parallel Track Transformers: Enabling Fast GPU Inference with Reduced Synchronization (arXiv 2602.07306)