Radiant AI Cloud Model Registry

What Radiant AI Cloud Model Registry is

Radiant AI Cloud Model Registry gives teams a place to organize custom AI models, keep versions, and prepare them for deployment. Model weights are uploaded using the RAIC command-line interface, with supported files including .safetensors, .json, .md, .txt, .py, .gitattributes, .yml, and LICENSE. Organizations begin with a default registry and can create more, selecting where weights are stored. When defining a model, users can indicate multiple preferred GPU types so the platform can match it with compatible GPUs available at that registry location.

A version passes through Uploaded and Synchronizing before reaching Available status, which makes it ready for deployment to Endpoints. Fine-tuning jobs feed into the registry automatically: when a job finishes, it creates a model and adds checkpoint versions ready for deployment. Fine-tuning guidance covers public Hugging Face datasets and custom .jsonl or .csv datasets with prompt and completion columns. RAIC uses Low-Rank Adaptation (LoRA) and FlashAttention-2 during training. The service supports deployment tracking and artifact storage alongside versioning. Its documentation also describes optional multi-factor authentication using an authenticator-app code after login.

Who it suits

It suits teams that need a shared workflow for storing, versioning, and deploying custom AI models, especially when model creation is tied to fine-tuning jobs. Teams can also use it to specify preferred GPU types and organize weights across registries.

Getting started

1. Visit the Radiant AI Cloud Model Registry documentation.
2. Create an organization to receive its default registry.
3. Create additional registries if needed and choose where weights will be stored.
4. Use the RAIC CLI to upload model weights in a supported file type.
5. Wait for a version to reach Available before deploying it to an Endpoint.

Verdict

Choose Radiant AI Cloud Model Registry if your team wants model versions, artifact storage, and a path from fine-tuning checkpoints to deployment. It is a paid service, and uploads use the RAIC CLI, so teams seeking a browser-only upload route should look elsewhere.