PAI-Model Gallery provides access to hundreds of pre-trained models from Hugging Face, ModelScope, and other leading communities. Deploy, fine-tune, and evaluate models through a no-code interface—no infrastructure setup required.
Key features
Model Gallery hosts pre-trained models across a wide range of AI domains: large language models (LLMs), AI-generated content (AIGC), computer vision (CV), natural language processing (NLP), and speech models.
One-click operations: Train, compress, evaluate, and deploy models in a few clicks.
Advanced customization: Customize hyperparameters and other configurations to match your workload requirements.
Filter models by source (ModelScope, PAI, or NIM) or by use case. Use cases span generative AI—large language models, multimodal large language models, text-to-image, and text-to-video—and computer vision tasks such as image classification and object detection. Each model card shows the model name, parameter count, description, and last updated date.
Workflow
Find a model: Browse and search for a model that fits your use case. For details, see Select a model for your business.
Deploy and debug: Deploy a pre-trained model as an online service, then debug it and evaluate inference results against your requirements. For details, see Deploy and debug a model.
Fine-tune a model: If the base model does not meet your performance requirements—for example, it lacks knowledge of your domain or output format—fine-tune it on your own dataset, then deploy the fine-tuned model for production. For details, see Model deployment and training.
Billing
Model Gallery itself is free. Training or deploying models through Model Gallery incurs resource charges. For pricing details, see Quick Start Billing.
Contact us
For product inquiries or support, join the PAI-Model Gallery user community on DingTalk. Scan one of the QR codes below, or search for Group 1 (79680024618) or Group 2 (77450028832) in the DingTalk app.
