🧑‍🔬 Jinpeng Li

🏛️ Affiliations

🏛️Chinese Academy of SciencesBeijing · China2026 · 2 paperscurrent🔬Computer Network Information CenterBeijing · China2026 · 2 paperscurrent🏢Huawei TechnologiesShenzhen · China2026 · 2 paperscurrent🎓University of Chinese Academy of SciencesBeijing · China2026 · 2 paperscurrent🎓Peking UniversityBeijing · China2026 · 1 papercurrent🎓Soochow UniversitySuzhou · China2026 · 1 papercurrent🎓South China University of TechnologyGuangzhou · China2026 · 1 papercurrent

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine.

🧭 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. StepHint: Multi-level Stepwise Hints Enhance Reinforcement Learning to Reason (2026 · 13 citations)
   Reinforcement learning with verifiable rewards (RLVR) improves LLM reasoning, but it has two weaknesses. The first is the near-miss reward problem: one early slip ruins an otherwise sound solution, so the rollout earns no reward. The second is exploration stagnation: the policy keeps sampling familiar solutions…
2. 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.
3. Pangu Pro MoE: Mixture of Grouped Experts for Efficient Sparsity (2025 · 4 citations)
   Sparse Mixture-of-Experts LLMs are cheap per token. However, when experts are spread across accelerators, uneven routing overloads some devices, and the slowest device limits overall speed. The Pangu Team at Huawei proposes Mixture of Grouped Experts (MoGE). Experts are split into equal, non-overlapping groups…
4. BOLT: Benchmarking Open-World Learning for Text Classification (2026 · 1 citation)
   Open-world text classification requires models to handle classes absent from training. Evaluation in this area has been fragmented. Most studies test on a few English intent datasets such as Banking and CLINC, and papers differ in how many classes are treated as known, how much data is labeled, and which metrics they…
5. Improved Policy Optimization for Mixture-of-Experts Models: Importance Sampling and Rewarding from an Expert-Centric Perspective (2026)
   Fine-tuning Mixture-of-Experts (MoE) LLMs with reinforcement learning is often unstable. The router can send the same token to different experts under the old and updated policies, so token-level importance sampling (IS) ratios, as used in GRPO, become noisy. GSPO computes the ratio per sequence, which is more stable…

All works

6 works of theirs in our library, the most cited first; 5 with a page here, the others link their original until we summarise them.

1. StepHint: Multi-level Stepwise Hints Enhance Reinforcement Learning to ReasonAnnual Meeting of the Association for Computational Linguistics (ACL)Kaiyi Zhang, Ang Lv, Jinpeng Li, et al.2026 · 13 citations
2. Pangu Embedded: An Efficient Dual-system LLM Reasoner with MetacognitionarXiv:2505.22375Hanting Chen, Yasheng Wang, Kai Han, et al.2025 · 13 citations
3. Pangu Pro MoE: Mixture of Grouped Experts for Efficient SparsityarXiv:2505.21411Yehui Tang, Xiaosong Li, Fangcheng Liu, et al.2025 · 4 citations
4. BOLT: Benchmarking Open-World Learning for Text ClassificationFindings of the Association for Computational LinguisticsChuan Qin, Xi Chen, Jinpeng Li, et al.2026 · 1 citation
5. Improved Policy Optimization for Mixture-of-Experts Models: Importance Sampling and Rewarding from an Expert-Centric PerspectiveFindings of the Association for Computational LinguisticsYining Qian, Jinpeng Li, Fei Mi, et al.2026
6. When to Trust Tools? Adaptive Tool Trust Calibration For Tool-Integrated Math Reasoning ↗Findings of the Association for Computational Linguistics: ACL 2026 · no summary yetRuotao Xu et al.2026

🤝 Co-authors here

12 of 24 co-authors with a page, the most papers together first.

• Fei Mi ×3
• Kai Han ×2
• Bin Wang
• Chuan Qin
• Dong Li
• Fangcheng Liu
• Feng Wang
• Hang Zhou
• Hanting Chen
• Haoyu Wang
• Hengshu Zhu
• Kaikai Song

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

• Findings of the Association for Computational Linguistics ×2: Findings of the Association for Computational Linguistics 2026 ×2
• Annual Meeting of the Association for Computational Linguistics (ACL): ACL 2026
• Findings of the Association for Computational Linguistics: ACL 2026

🔗 Cited by

Papers of our library cite their works 29 times (of the 149,439 papers whose bibliographies we read); 11 citing papers with a page here →.

Updated 1 November 2026

↑ back to top

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