Skip to navigation
ResourcesPlatform v2 (Legacy)Deployment & ArchitectureSnowflake Native Application

Snowflake SPCS Setup

View as Markdown

This guide explains how to use the Kumo Fine-Tune SDK with the Kumo Snowflake Native Application running on Snowpark Container Services (SPCS).

Running SDK Outside Snowflake

When running the SDK outside Snowflake, initialize it with your Snowflake credentials. The SDK supports both password-based and key-pair authentication:

import kumoai as kumo
# Option 1: password-based authentication
credentials = {
    "user": "<your username>",
    "password": "<your password>",
    "account": "<your account>",
}
kumo.init(url="https://<kumo_spcs_deployment_url>/api", snowflake_credentials=credentials)
# Option 2: key-pair authentication (same key format as connector auth in the UI)
credentials = {
    "user": "<your username>",
    "private_key": "<PEM-encoded private key>",
    "account": "<your account>",
}
# Optional: for encrypted keys
# credentials["private_key_passphrase"] = "<passphrase>"
kumo.init(url="https://<kumo_spcs_deployment_url>/api", snowflake_credentials=credentials)

For detailed examples of using the SDK outside Snowflake, refer to Introduction by Example.

Running SDK on Snowflake

To run the SDK directly on Snowflake, use a Snowflake notebook configured for Snowpark Container (choose Run on Container). This setup provides a managed environment for running machine learning workloads.

Snowflake Notebook Configuration

For instructions on setting up your notebook environment, refer to the Snowflake ML Notebooks on SPCS documentation.

Using External Access

If your notebook environment has external access configured (refer to External Access Configuration), you can install the Kumo Fine-Tune SDK directly using pip:

pip install kumoai==<kumo_app_version>

Using Package Installation

If external access is not available or permitted in your environment, follow these steps:

  1. Contact the Kumo team to obtain the SDK package zip file
  2. Upload the package to your Snowflake stage
  3. Install the package from the stage

The recommended way to upload the package is using the Snowsight UI:

Package Installation Flow

  1. Sign in to Snowsight
  2. Select Data » Add Data
  3. On the Add Data page, select Load files into a Stage
  4. In the Upload Your Files dialog:
    • Select the kumoai.zip file (maximum file size: 250 MB)
    • Select the database and schema where you want to create the stage
    • Select or create your stage (for example, ‘kumo_stage’)
    • Optionally, specify a path within the stage
  5. Select Upload

For detailed instructions, refer to:

Getting Started

When running inside a Snowflake notebook, the SDK automatically uses the Snowflake session context: it inherits your notebook’s active credentials and connection. Initialize with just the application name:

import kumoai as kumo
# Initialize with just the application name - no additional credentials needed
kumo.init(snowflake_application='YOUR_APP_NAME')
# Create a Snowflake connector with your Snowflake resources
my_connector = kumo.SnowflakeConnector(
    name="YOUR_CONNECTOR_NAME",
    account="YOUR_SNOWFLAKE_ACCOUNT_NAME",
    warehouse="YOUR_SNOWFLAKE_WAREHOUSE_NAME",
    database="YOUR_SNOWFLAKE_DATABASE_NAME",
    schema_name="YOUR_SNOWFLAKE_SCHEMA_NAME",
)
# Start using Kumo's features with your connector
# ...

For graph and training setup, refer to Introduction by Example.

Additional Resources