Yichang Zhang

Timeline

1. 2019

   A knowledge-based dialog system

   The paper "Towards Knowledge-Based Recommender Dialog System" appears in 2019 with Yichang Zhang among its authors, alongside Qibin Chen, Junyang Lin, Ming Ding, Yukuo Cen, Hongxia Yang and Jie Tang. It proposes KBRD, an end-to-end framework that integrates a recommender system with a dialog generation system.

2. 2020

   InterBERT multimodal pretraining

   In 2020, Zhang co-authors InterBERT with Junyang Lin, An Yang, Jie Liu, Jingren Zhou and Hongxia Yang, the first model of the M6 series of multimodal pretraining methods. The Chinese InterBERT is pretrained on 3.1M image-text pairs from mobile Taobao and is deployed online for topic-based recommendation.

3. 2021

   M6, a Chinese multimodal pretrainer

   In 2021, Zhang is among the authors of M6, a Chinese multimodal pretrainer built on a dataset of over 1.9TB of images and 292GB of texts. The authors scale the model to 10 billion and 100 billion parameters and build the largest pretrained model in Chinese.

4. 2022

   Chinese CLIP models

   In 2022, Zhang co-authors Chinese CLIP with An Yang, Junshu Pan, Junyang Lin, Rui Men, Jingren Zhou and Chang Zhou. The team builds a large-scale dataset of Chinese image-text pairs and develops 5 Chinese CLIP models of 77 to 958 million parameters.

5. 2023

   The Qwen technical report

   In September 2023, Zhang is among the authors of the Qwen technical report, which introduces Qwen as the first installment of a large language model series. The series has base pretrained models, the Qwen-Chat chat models, and the coding models Code-Qwen and Code-Qwen-Chat.

6. 2024

   The Qwen2.5 technical report

   In December 2024, Zhang is among the authors of the Qwen2.5 technical report. Qwen2.5 scales its pre-training data from 7 trillion to 18 trillion tokens and adds supervised finetuning with over 1 million samples and multistage reinforcement learning.

7. 2025

   Qwen3, Qwen-Image and Qwen3Guard

   In May 2025, Zhang is among the authors of the Qwen3 technical report for the series developed by the Qwen team at Alibaba Cloud. Qwen3 comprises dense and Mixture-of-Expert models from 0.6 to 235 billion parameters and expands multilingual support from 29 to 119 languages and dialects. Zhang also co-authors the Qwen-Image technical report in August 2025, on an image generation foundation model, and the Qwen3Guard technical report in October 2025, on a series of multilingual safety guardrail models.

8. As of October 2026

   On the Qwen Team

   As of October 2026, Zhang is with the Qwen Team at Alibaba Group. Their research interests include NLP, reinforcement learning, deep learning, machine learning and artificial intelligence.