Scholar
Xin Dong
Google Scholar ID: O8nBN64AAAAJ
Research Scientist & Tech Lead, NVIDIA Research; Harvard University
Citations & Impact
Citations
2,262
H-index
20
i10-index
26
Publications
20
Co-authors
8
list available
Contact
Publications
Resume (English only)
Academic Achievements
- - Released a new series of NVIDIA Nemotron models, aiming to contribute to the journey of building the next-generation architecture
- - Significantly scaled up the compute (3k steps) and data (5 distinct domains) for RL, setting a new scaling record for open-source RL training
- - Released CLIMB, a robust LLM pre- and post-training dataset building method
- - Released the first intra-layer (head-wise fusion) hybrid model, Hymba-1.5B, accepted by ICLR 2025 as a spotlight
- - Additive Power-of-Two Quantization (ICLR'20) is now supported in official PyTorch
Research Experience
- - Research Scientist at NVIDIA Research, 2024 - Present
- - Research Scientist at Sony, 2023 - 2024
- - Research Scientist Intern at Meta Reality Lab, 2021, 2022
- - Research Scientist Intern at NVIDIA, 2020
- - Research Scientist Intern at Tencent America, 2019
Education
- - Degree: Ph.D.
- - School: Harvard University
- - Advisor: H.T. Kung
- - Time: 2023
- - Major: Computer Science
- - Honor: Recipient of the Harvard James Mills Peirce Fellowship
Background
- - Research Interests: Training of intelligent and efficient foundation models
- - Field: Computer Science
- - Brief Introduction: A research scientist at NVIDIA Research, leading the training of intelligent and efficient foundation models.
Miscellany
- - Personal Interests: Not mentioned
