1. Industry AI Model Platform / 
  2. v1.x / 
  3. Resource Management / 
  4. Model Inference / 
  5. Public Inference Overview

Documentation

Last updated:
    Edit

    Public Inference Overview

    What is Public Inference

    Public inference is a shared entry point provisioned by administrators. Users do not create their own instance: open Model Inference → Public Inference, pick a model, chat in the Playground, or call the API.

    The list shows two kinds of public models:

    Type Access Description
    Local deployment In-cluster inference An administrator deploys a platform repository as a shared service. The card shows the runtime framework and resource specification
    Cloud API Unified gateway An administrator registers an external vendor channel (for example Moonshot, DeepSeek, or Alibaba DashScope). The card shows the vendor; it does not consume cluster GPUs

    Both types expose an OpenAI-compatible API. Cloud models are forwarded through /aigateway (or /platform-model/aigateway) on the console. The request model field is the platform model ID.

    Compared with Dedicated Instances

    Public inference Dedicated instance
    Who creates it Platform administrator Regular user
    Compute Shared local deployment, or the cloud vendor’s compute Exclusive specification chosen by the user
    Entry Model Inference → Public Inference Model Inference → Dedicated Instances, or Deploy on the model detail page
    Auth Cloud channels always require an access token; local deployments follow the public / private setting Chosen when the instance is created

    Use a dedicated instance when you need exclusive resources, custom engine args, or a specific quantization.