DataWorks large model services-DataWorks(DataWorks)-阿里云帮助中心

DataWorks large model services let you deploy models on serverless resource groups and call them directly in data development tasks. All traffic stays in a private network through PrivateLink, keeping data secure.

Supported models

Category	Model	Description
Large language model	Qwen3-32B	Qwen3 is the latest generation of large language models in the Qwen series, offering a full set of dense and Mixture-of-Experts (MoE) models with breakthrough capabilities in reasoning, instruction following, agent functions, and multilingual support. Qwen3. Important Deploying the `0.6B`, `1.7B`, `4B`, and `8B` models requires a minimum of `24 GB` of GPU memory. Deploying the `14B` model requires a minimum of `48 GB` of GPU memory. Deploying the `32B` model requires a minimum of `96 GB` of GPU memory.
	Qwen3-14B
	Qwen3-8B
	Qwen3-4B
	Qwen3-1.7B
	Qwen3-0.6B
	Qwen3-Embedding-8B
	Qwen3-Embedding-4B
	Qwen3-Embedding-0.6B
	DeepSeek-R1-0528-Qwen3-8B	DeepSeek extensively uses reinforcement learning, significantly improving inference capabilities with minimal annotated data. DeepSeek.
	DeepSeek-R1-Distill-Qwen-1.5B
	DeepSeek-R1-Distill-Qwen-7B
	DeepSeek-R1-Distill-Qwen-14B
Vector model	BGE-M3	BGE-M3 is a general-purpose vector model that supports dense, multi-vector, and sparse retrieval. It handles inputs up to 8,192 tokens and supports over 100 natural languages.
Vector model	BGE-Large-zh-v1.5	BGE Embedding is a general-purpose vector model pre-trained with RetroMAE and trained on large-scale paired data using contrastive learning.

Features of large model services

Fast deployment

A guided visual interface enables low-code model deployment. Create and publish model services with minimal configuration — no infrastructure code required.

Secure internal communication

Models are deployed in fully managed DataWorks resource groups and use PrivateLink and private zone for cross-VPC communication with network isolation. Access the model from your VPC internal network.

Network-isolated communication: PrivateLink creates a dedicated channel between your VPC and the DataWorks resource group VPC, keeping traffic off the public internet.
Private domain name resolution: A private zone provides internal DNS that automatically forwards domain name requests to the model service, enabling domain-based access from your VPC.

Automated connection

When you deploy a model service in a DataWorks resource group or configure a VPC for it, the system automatically:

Establishes a cross-VPC connection: Creates a PrivateLink endpoint in your VPC and sets up an encrypted channel to the DataWorks resource group VPC.
Configures domain name resolution: Adds DNS forwarding rules in your VPC to route requests to the DataWorks model service.

Model invocation flow

When you call a model service using a domain name:

Request parsing: The private zone resolves DNS requests from your VPC to the private IP of the DataWorks resource group VPC.
Traffic forwarding: PrivateLink transmits the request securely from your VPC to the DataWorks resource group VPC.
Service response: The model instance processes the request and returns the inference result.

The entire process runs in a private network with no public IP exposure.

Model deployment and usage

Select and deploy models for use in data development tasks: