Model Repositories
TIR Model Repositories give you a central, scalable place to store your AI/ML model weights and configuration files. They run on E2E Object Storage (EOS) with S3-compatible APIs, so it's easy to version models, collaborate with your team, and connect them to TIR inference.
Quick Start
Quick Start Guide
Create and use your first model repository on TIR.
Features
Explore model storage, versioning, and repository capabilities.
Plans & Pricing
Review pricing and storage billing details for model repositories.
FAQs
Find troubleshooting help and answers to common questions about model repositories.
What are Model Repositories?
Model Repositories are managed storage systems designed specifically for storing and organizing AI/ML model artifacts. They serve as the foundation for model deployment workflows, allowing you to:
Store model weights and configuration files in a centralized location
Version models using flexible folder structures (v1, v2, etc.)
Share models across team members within a project
Integrate seamlessly with Model Endpoints for automated model loading
Access models from Instances and Inference services
Key Characteristics
Structure
Flexible Model Definition
A model in TIR is simply a directory on EOS. There is no rigid structure, format, or framework required. Use subfolders like v1 and v2 to track versions.
API
S3-Compatible Storage
Built on E2E Object Storage (EOS) with full S3 API support. Use standard tools like Minio CLI (mc) and s3cmd without vendor lock-in.
Flexibility
Multiple Storage Options
Use a new EOS bucket, link an existing EOS bucket, or connect an external S3-compatible bucket for your model repository.
Use Cases
Model Versioning and Management
- Store multiple versions of the same model (v1, v2, production, staging)
- Track model iterations and roll back to previous versions
- Maintain model lineage and metadata
Model Deployment Pipeline
- Store models before deploying to Model Endpoints
- Support CI/CD workflows for model deployment
Fine-tuning Workflows
- Store base models for fine-tuning
- Save fine-tuned model checkpoints
- Download models to Instances for training
Multi-Framework Support
- Store PyTorch models (.pth, .pt files)
- Store TensorFlow models (SavedModel format)
- Store Triton model repositories
- Store custom model formats
Direct Deployment from Model Repository
- Use the Deploy Model option in the Model Repository table to deploy as a Model Endpoint
- Select a framework (for example, vLLM, SGLang, NVIDIA Triton) and link the repository to the endpoint
API Reference
Model Repository API Reference
Programmatically create, list, and delete model repositories in TIR.
/teams/{Team_Id}/projects/{Project_Id}/serving/model/List model repositories/teams/{Team_Id}/projects/{Project_Id}/serving/model/Create model repository/teams/{Team_Id}/projects/{Project_Id}/serving/model/model_types/Get model types/teams/{Team_Id}/projects/{Project_Id}/serving/model/{Model_repo_id}/Delete model repository