Elamur Research⌕
☰
home / authors / Renjie Pi

Renjie Pi

Renjie Pi has 6 papers in our library, in 2024–2026, most of them on Reasoning, Language models & pretraining, Bias, fairness & safety; often with Jipeng Zhang, Rui Pan, Hanze Dong.

6 papers in our library · 2024–2026

OpenAlex · ORCID · h-index 14 · 70 works · 552 citations OpenAlex, 2026-10-14

📄 6 works in our library · 🔗 Cited 258 times in our library · 3 more in our library without a page yet · ⭐ Paper Stars ranking: #5779

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: Google Scholar · Semantic Scholar · dblp · ACL Anthology

 Affiliations

🎓Hong Kong University of Science and TechnologyHong Kong · China2021–2026 · 53 papersEducationcurrent

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 1 more institution on single bylines are not listed.

 Fields, topics and concepts

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

📚 Shelves 8The field of science this work belongs to (OpenAlex fields).Computer Science · 6 worksArtificial Intelligence · 5 worksArts and Humanities · 3 worksLanguage and Linguistics · 3 worksComputer Vision and Pattern Recognition · 2 worksComputational Theory and Mathematics · 1 workSocial Sciences · 1 workSociology and Political Science · 1 work

🧭 Topics 4Research subjects and languages the work is about.Bias, fairness & safety · 1 workReasoning · 2 worksEfficient models & compression · 1 workLanguage models & pretraining · 1 work

 Works by year

2020s

20262 papers20252 papers20242 papers

 Papers

The most cited first (OpenAlex and our library).

  1. Mitigating the Alignment Tax of RLHF (2024 · 83 citations)
    RLHF makes LLMs follow human preferences, but it also tends to erode skills learned in pre-training. This loss is called the alignment tax. The authors measured it by aligning OpenLLaMA-3B on the Anthropic helpfulness/harmlessness data with Rejection Sampling Fine-tuning (RSF) and DPO. They then tracked performance…
  2. MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance (2024 · 58 citations)
    Multimodal large language models such as LLaVA can be steered into harmful output by malicious images, even when the same request in plain text would be refused.
  3. MA-LoT: Model-Collaboration Lean-based Long Chain-of-Thought Reasoning enhances Formal Theorem Proving (2025 · 18 citations)
    The paper addresses automated theorem proving in Lean4. Current LLM approaches fall into two camps. Step-by-step tree search uses verifier feedback but is costly and lacks high-level planning. Whole-proof generation exploits natural-language reasoning but never sees intermediate verifier signals. MA-LoT bridges these…
  4. Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation (2026 · 16 citations)
    Post-training a single LLM across many reasoning and agentic domains is hard. Mixing diverse RL environments can destabilize training, and gains in one area often erode others. Nemotron-Cascade 2 extends NVIDIA's earlier Cascade RL approach, which trains domain by domain in sequence, to address this. The base model…
  5. VLM2-Bench: A Closer Look at How Well VLMs Implicitly Link Explicit Matching Visual Cues (2025 · 12 citations)
    The paper asks whether vision-language models can do something people do routinely: recognise that the same object or person appears in several images or video frames by comparing visual traits, without knowing who or what it is.
  6. LongVideoAgent: Multi-Agent Reasoning with Long Videos (2026 · 3 citations)
    Answering questions about hour-long videos is difficult because the relevant evidence is scattered and sparse.

All works

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

  1. Mitigating the Alignment Tax of RLHFConference on Empirical Methods in Natural Language Processing (EMNLP)Yong Lin, Hangyu Lin, Wei Xiong, et al.2024 · 83 citations
  2. DetGPT: Detect What You Need via Reasoning ↗Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · no summary yetRenjie Pi et al.2023 · 83 citations
  3. G-LLaVA: Solving Geometric Problem with Multi-Modal Large Language Model ↗arXiv:2312.11370 · no summary yetJiahui Gao et al.2023 · 76 citations
  4. MLLM-Protector: Ensuring MLLM's Safety without Hurting PerformanceConference on Empirical Methods in Natural Language Processing (EMNLP)Renjie Pi, Tianyang Han, Jianshu Zhang, et al.2024 · 58 citations
  5. TheoremLlama: Transforming General-Purpose LLMs into Lean4 Experts ↗arXiv:2407.03203 · no summary yetRuida Wang et al.2024 · 51 citations
  6. MA-LoT: Model-Collaboration Lean-based Long Chain-of-Thought Reasoning enhances Formal Theorem ProvingarXiv:2503.03205Ruida Wang, Rui Pan, Yuxin Li, et al.2025 · 18 citations
  7. Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy DistillationarXiv:2603.19220Zhuolin Yang, Zihan Liu, Yang Chen, et al.2026 · 16 citations
  8. VLM2-Bench: A Closer Look at How Well VLMs Implicitly Link Explicit Matching Visual CuesAnnual Meeting of the Association for Computational Linguistics (ACL)Jianshu Zhang, Dongyu Yao, Renjie Pi, et al.2025 · 12 citations
  9. LongVideoAgent: Multi-Agent Reasoning with Long VideosAnnual Meeting of the Association for Computational Linguistics (ACL)Runtao Liu, Ziyi Liu, Jiaqi Tang, et al.2026 · 3 citations

 Co-authors here

12 of 31 co-authors with a page, the most papers together first.

Venues

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

 Cited by

Papers of our library cite their works 258 times (of the 149,439 papers whose bibliographies we read); 57 citing papers with a page here →.

Updated

↑ back to top

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