🧑‍🔬 Dong Li

🏛️ Affiliations

🎓Capital Medical UniversityBeijing · China2009–2026 · 98 paperscurrent🏥Beijing Tongren HospitalBeijing · China2016–2026 · 79 paperscurrent🏛️Ministry of Industry and Information TechnologyBeijing · China2020–2026 · 54 paperscurrent🏥Beijing Anzhen HospitalBeijing · China2009–2025 · 15 papers🏥Tianjin Medical University General HospitalTianjin · China2008–2025 · 9 papers🏥First Affiliated Hospital of Harbin Medical UniversityHarbin · China2012–2022 · 8 papers🎓Harbin Medical UniversityHarbin · China2012–2022 · 8 papers🤝The Lundquist InstituteTorrance · United States2013–2022 · 7 papers🎓Tianjin Medical UniversityTianjin · China2008–2025 · 6 papers🏥Harbor–UCLA Medical CenterTorrance · United States2014–2023 · 6 papers

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 30 more institutions on single bylines are not listed.

🧭 Fields, topics and concepts

Of its works in our library, with the number of works on each.

📅 Works by year

2020s

📄 Papers

The most cited first (OpenAlex and our library).

1. ZeRO-Offload: Democratizing Billion-Scale Model Training (2021 · 309 citations)
   Training models with billions of parameters usually needs many GPUs, because existing parallel methods only work when the combined GPU memory can hold the parameters, gradients and optimizer states.
2. MiniMax-01: Scaling Foundation Models with Lightning Attention (2025 · 50 citations)
   Most current LLMs and VLMs handle 32K–256K tokens of context. That is often too little for tasks like reading whole books, working across entire code repositories, or many-shot in-context learning. The main obstacle is that softmax attention cost grows quadratically with length. MiniMax-01 addresses this with a…
3. TransNormerLLM: A Faster and Better Large Language Model with Improved TransNormer (2023 · 33 citations)
   Linear attention promises to remove the quadratic cost of softmax attention. In practice, earlier linear-attention models have trailed Transformers in language-modeling quality and rarely ran faster on real hardware. The authors close this gap with TransNormerLLM, a redesigned version of their TransNormer…
4. Lightning Attention-2: A Free Lunch for Handling Unlimited Sequence Lengths in Large Language Models (2024 · 31 citations)
   Linear attention should cost O(n) instead of the O(n²) of softmax attention. In causal language modeling, however, existing implementations lose this advantage: the kernel trick needs a cumulative summation (cumsum) that parallelizes poorly on GPUs. Lightning Attention-1 had already reduced the memory-I/O bottleneck…
5. Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognition (2025 · 13 citations)
   Reasoning-focused LLMs such as DeepSeek-R1 are expensive to deploy because they are very large and generate long chains of thought, sometimes tens of thousands of tokens per problem.
6. Pangu Ultra: Pushing the Limits of Dense Large Language Models on Ascend NPUs (2025 · 6 citations)
   Pangu Ultra is Huawei's effort to show that a very deep dense Transformer, trained entirely on Ascend NPUs instead of GPUs, can compete with much larger sparse MoE systems.

🤝 Co-authors here

12 of 38 co-authors with a page, the most papers together first.

• Kaikai Song ×2
• Weigao Sun ×2
• Weixuan Sun ×2
• Xuyang Shen ×2
• Yasheng Wang ×2
• Yiran Zhong ×2
• Zhen Qin ×2
• Aonian Li
• Bangwei Gong
• Bin Wang
• Bo Yang
• Boji Shan

🔗 Cited by

Papers of our library cite their works 378 times (of the 149,439 papers whose bibliographies we read); 53 citing papers with a page here →.

Updated 30 October 2026

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/authors/dong-li/