One‑Click Deployment of Baidu Qwen3 Large Models on Baidu Baige AI Platform

The Qwen3 series comprises eight models ranging from 0.6 billion to 235 billion parameters, including two Mixture‑of‑Experts (MoE) models and six dense models, all supporting a mixed‑inference mechanism.

Baidu Baige AI Heterogeneous Computing Platform now offers a one‑click deployment solution for the entire Qwen3 family, providing enterprises with a fast, stable, and cost‑effective way to bring large models into production.

Deployment Steps

Log in to the Baidu Baige AI platform and navigate to the Quick Start section to locate the Qwen3 models.

Click the model card’s One‑Click Deploy button. The platform supports SGLang and vLLM as inference acceleration back‑ends.

The system recommends the minimum resource configuration for each model; you can adjust the settings as needed. Note that you must purchase compute resources in advance and create either a self‑managed or fully‑managed resource pool on the platform.

After deployment succeeds, open the Online Services list to view service details, obtain the endpoint URL and access token, and start invoking the model.

The Baidu Baige platform also supports advanced PD‑separated inference deployment strategies, including adaptive PD ratios, fine‑grained PD load balancing, optimal hybrid parallelism, and dynamic redundant expert orchestration. These features reduce total‑pipeline‑operation‑time (TPOT) by 40 % and inference cost by 95 %.

This solution powers Baidu Cloud’s Qianfan platform, serving over 400 k customers. Since its launch, inference throughput has increased by 20× and overall speed has improved by more than 50 %.