Achieving Human Parity on Automatic Chinese to English News Translation
Technical Report 2018
LLM Post-Training · Reasoning · Multilingual · Multimodal
Zhirui Zhang
Researcher and builder across large-model systems, translation, multimodal capability, and deployable intelligence.
Most recently: AI Technical Advisor and entrepreneurial partner at IDEA Research.
Profile
About
I focus on post-training and reasoning models for large language models, while continuing to build on a long research track in multilingual NLP, machine translation, speech translation, and dialogue systems. Across industry labs and product teams, I have worked on frontier-scale pretraining, general post-training, reasoning-oriented optimization, translation and multilingual capability building, and practical deployment in user-facing systems.
I received my Ph.D. from the University of Science and Technology of China through a joint training program with Microsoft Research Asia. My recent work bridges academic research and real-world model development, with a sustained interest in reliable, iterative, and deployable large-model systems.
Current Focus
Research Areas
LLM Post-Training Reasoning Models Multilingual Modeling Machine Translation Multimodal Capability Building Production Model Systems
Selected Work
Representative Papers
Influential
Document-Level Machine Translation with Large Language Models
EMNLP 2023
Adaptive Nearest Neighbor Machine Translation
ACL-IJCNLP 2021
Regularizing Neural Machine Translation by Target-bidirectional Agreement
AAAI 2019
Incorporating BERT into Parallel Sequence Decoding with Adapters
NeurIPS 2020
Simple and Scalable Nearest Neighbor Machine Translation
ICLR 2023
Recent
SWE-AGI: Benchmarking Specification-Driven Software Construction with MoonBit in the Era of Autonomous Agents
arXiv 2026
Is PRM Necessary? Problem-Solving RL Implicitly Induces PRM Capability in LLMs
NeurIPS 2025
Safe Delta: Consistently Preserving Safety when Fine-Tuning LLMs on Diverse Datasets
ICML 2025
Simple o3: Towards Interleaved Vision-Language Reasoning
arXiv 2025
Lost in the Source Language: How Large Language Models Evaluate the Quality of Machine Translation
ACL 2024 Findings
Experience
Concise Timeline
2025.12 - 2026.3
IDEA Research
AI Technical Advisor and entrepreneurial partner working on model-and-tool systems for MoonBit and more reliable code intelligence.
2024.3 - 2025.9
Huawei
Technical expert leading general post-training, reasoning-model exploration, multilingual capability building, and practical large-model deployment.
2023.11 - 2024.3
StepFun
Algorithm expert contributing to trillion-parameter MoE pretraining, FP8 language-model exploration, and long-context capability validation.
2021.9 - 2023.11
Tencent AI Lab
Senior researcher building translation training platforms, interactive translation models, personalized MT, and multilingual research systems.
2019.7 - 2021.8
Alibaba DAMO Academy
Algorithm expert for multilingual translation, speech translation, automated training pipelines, and commercial translation services.
2015.7 - 2019.6
Microsoft Research
Research intern across MSRA and Redmond, working on neural machine translation, dialogue systems, and controllable text generation.
Service & Recognition
Academic Service and Awards
Service
Long-term reviewer or program committee member for ACL, EMNLP, NAACL, AAAI, IJCAI, NeurIPS, ICML, and ICLR, with prior service as an ACL area chair.
Awards
- National Scholarship
- Google Scholarship
- MSRA Star of Tomorrow Internship Award of Excellence
Writing / Notes
Short technical notes beyond papers.
I am setting up a lightweight blog for notes on post-training, translation, multimodal systems, and practical lessons from building deployable model stacks.
Models Translation Multimodal Systems
Contact