Deploy and Use Qwen3‑Coder on Alibaba Cloud PAI for AI‑Powered Coding

Related Background

Qwen3‑Coder, the new AI programming large model from Alibaba, has been open‑sourced. It offers strong code generation and agent capabilities, with the 480B‑A35B‑Instruct MoE variant requiring at least 16×96 GB GPU memory for inference.

Qwen3‑Coder x PAI‑DSW

Using Qwen3‑Coder in the interactive modeling environment PAI‑DSW solves slow environment debugging and training code failures. Enterprise customers can deploy the model as a service for high‑performance needs, while individual developers can use the free plan launched on July 26, with unlimited token usage in IDEs such as Tongyi Lingma, VSCode, and JetBrains.

PAI‑DSW Introduction

PAI‑DSW (Data Science Workshop) provides a cloud‑based AI development IDE with pre‑installed environments, support for OSS/NAS/CPFS datasets, and a variety of open‑source framework images, enabling efficient model development and lifecycle management.

Specific Steps

Step 1: Create a DSW Instance

Log in to Alibaba Cloud, open the AI platform PAI workspace, and select “Model Development and Training → Interactive Modeling (DSW)”. Create a new DSW instance and choose the image modelscope:1.28.0-pytorch2.6.0-gpu-py311-cu124-ubuntu22.04, which includes the required Node.js version.

In the custom startup script configure:

export NVM_DIR="$HOME/.nvm"
[ -s "$NVM_DIR/nvm.sh" ] && . "$NVM_DIR/nvm.sh"
nvm use 22
npm i -g @qwen-code/qwen-code

Set the model invocation information in the environment variables after deploying the model via PAI‑Model Gallery (see linked guide).

Step 2: Enter Agentic CLI Qwen Code

When the DSW instance is running, open the terminal and type qwen to launch the interactive Agentic command‑line tool, allowing natural‑language control of Qwen3‑Coder.

Step 3: Generate a Notebook Tutorial

Provide a prompt such as “Write a notebook tutorial for SFT based on Qwen3‑Coder, first check GPU resources, ensure the notebook runs quickly, and obtain the model from ModelScope.” Qwen3‑Coder will produce an executable ipynb file, which can be run in JupyterLab.

Conclusion

Alibaba Cloud PAI offers a one‑stop AI development platform covering data labeling, model building, training, deployment, and inference optimization. With free, token‑less access to Qwen3‑Coder through DSW WebIDE or IDE plugins, developers can accelerate AI‑assisted coding workflows.