ERNIE-4.5-21B-A3B-Thinking: Lightweight Model Reasoning Capabilities Upgraded

1. Tutorial Introduction

ERNIE-4.5-21B-A3B-Thinking is a lightweight reasoning model "Thinking Edition" released by the Baidu Wenxin Yiyan team in September 2025. ERNIE-4.5-21B-A3B-Thinking adopts a mixture of experts (MoE) architecture with a total parameter size of 21B. Each token activates 3B parameters and is trained through instruction fine-tuning and reinforcement learning. ERNIE-4.5-21B-A3B-Thinking is a deep thinking model trained on the basis of ERNIE-4.5-21B-A3B. It supports a context window of 128K and is suitable for complex reasoning tasks that require long context. This model not only achieves significant improvements in tasks that require human experts such as logical reasoning, mathematics, science, code and text generation, but also has efficient tool calling capabilities and can support the automated processing of complex tasks. The relevant paper results are "ERNIE4.5 Technical Report".

The computing resources used in this tutorial are a single RTX A6000 card.

2. Effect display

3. Operation steps

1. Start the container

If "Model" is not displayed, it means the model is being initialized. Since the model is large, please wait about 3-4 minutes and refresh the page.

2. Usage steps

4. Discussion

🖌️ If you see a high-quality project, please leave a message in the background to recommend it! In addition, we have also established a tutorial exchange group. Welcome friends to scan the QR code and remark [SD Tutorial] to join the group to discuss various technical issues and share application effects↓

Citation Information

The citation information for this project is as follows:

@misc{ernie2025technicalreport,
      title={ERNIE 4.5 Technical Report},
      author={Baidu-ERNIE-Team},
      year={2025},
      primaryClass={cs.CL},
      howpublished={\url{https://ernie.baidu.com/blog/publication/ERNIE_Technical_Report.pdf}}
}