Skip to main content

Replicate

Replicate is a platform for running machine learning models in the cloud. The Replicate node in Boltic lets you connect your Replicate account and automate workflows involving AI models, predictions, fine-tuning, and deployments.

Replicate Account​

  • Select the Replicate account you want to connect.
  • If you don't have an existing account, add one using the Add New Account option before proceeding.
  • Your Replicate credentials are encrypted and can be removed at any time.

Available Resources​

Here are the available resources you can use with the Replicate node:

Resources​

  • Models: Work with public Replicate models — retrieve, create, update, delete models, run predictions, fetch READMEs, and list examples.
  • Model Versions: Manage versions of a model — list versions, retrieve a specific version, and delete a version.
  • Predictions: Run models and manage predictions — list, retrieve, create, and cancel predictions.
  • Trainings: Create and manage model fine-tuning jobs — create, list, retrieve, and cancel trainings.
  • Collections: Browse curated groups of Replicate models — list collections and retrieve a collection with its models.
  • Deployments: Create and manage dedicated model deployments — create, list, retrieve, update, delete deployments, and run predictions via a deployment.
  • Hardware: List available hardware SKUs for running and deploying models.
  • Search: Search public models, collections, and docs using a text query (beta).
  • Custom API: Execute custom authenticated Replicate API calls with full flexibility.

Basic Operations​

Here are the basic operations you can perform with the Replicate node:

Models​

  • Get a Model: Retrieve a single model by owner and name.
  • Create a Model: Create a new model under your account with hardware, visibility, and metadata.
  • Update Model Metadata: Update description, README, URLs, and cover image on an existing model.
  • Delete a Model: Delete a private model that has no versions.
  • Create Prediction Using Model: Run an official model directly without specifying a version.
  • Get Model README: Retrieve the README for a model in Markdown format.
  • List Model Examples: Retrieve a list of example predictions for a model.
  • Search Public Models: Search public models using a plain-text query (uses the non-standard QUERY HTTP method).

Model Versions​

  • List Model Versions: Retrieve a list of available versions for a model.
  • Get a Model Version: Retrieve details for a specific model version by ID.
  • Delete a Model Version: Delete a single owned model version by ID.

Predictions​

  • List Predictions: Retrieve a paginated list of predictions you have created.
  • Get a Prediction: Retrieve the status and output of a single prediction.
  • Create a Prediction: Run a model version and create a new prediction.
  • Cancel a Prediction: Cancel a prediction that is currently running.

Trainings​

  • Create a Training: Create a new fine-tuning training job using a model version.
  • List Trainings: Retrieve a paginated list of trainings you have created.
  • Get a Training: Retrieve details and status of a single training by ID.
  • Cancel a Training: Cancel a training that is currently running.

Collections​

  • List Collections: Retrieve a paginated list of curated model collections.
  • Get a Collection: Retrieve a single collection including all its models.

Deployments​

  • Create a Deployment: Create a new deployment for a model version with hardware and scaling configuration.
  • List Deployments: Retrieve a paginated list of your deployments.
  • Get a Deployment: Retrieve details for a specific deployment by owner and name.
  • Update a Deployment: Update hardware, scaling, or model version of an existing deployment.
  • Delete a Deployment: Delete a deployment that has been offline for 15 or more minutes.
  • Create Prediction via Deployment: Run a model using a specific named deployment.

Hardware​

  • List Available Hardware: Retrieve a list of available hardware SKUs for running and deploying models.
  • Search Models, Collections, and Docs: Search public models, collections, and documentation using a text query. Returns results with AI-generated descriptions, tags, and relevance scores. Limit results between 1 and 50 (default 20). This endpoint is currently in beta and may change.

Custom Operations​

  • Custom API Request: Make a custom authenticated request to any Replicate API endpoint.

Conclusion​

Replicate helps you automate AI model workflows inside Boltic. Use the Replicate node to manage models, versions, predictions, trainings, collections, deployments, hardware, search, and custom API requests in your workflows.

If you need further assistance, contact Boltic support at [email protected].