
<section>
    <h1>Cortex Inference</h1>
</section>


Deprecated in favor of "max_tokens", which has identical behavior.

The maximum number of output tokens to produce. The default value is model-dependent.

The model name. See documentation for possible values.

The provisioned throughput ID to be used with the request.

An object describing response format config for structured-output mode.

Temperature controls the amount of randomness used in response generation. A higher temperature corresponds to more randomness.

Configures how tools should be selected and used during the interaction.
Controls whether tool use is automatic, required, or specific tools should be used.


List of tools to be used during tool calling

Defines a tool that can be used by the agent.
Tools provide specific capabilities like data analysis, search, or generic functions.


Threshold probability for nucleus sampling. A higher top-p value increases the diversity of tokens that the model considers, while a lower value results in more predictable output.

The response when the guardrails model marks the completion as unsafe

GuardrailsConfig

Text-completion response for non-streaming request.

Usage

Server-sent events for streaming text-completion updates.

Streaming text-completion response event.

common-cortex-tool.yaml#/components/schemas/StreamingTextContent

Configure External Oauth with Snowflake (see <a href=https://docs.snowflake.com/en/user-guide/oauth-ext-overview>External OAuth overview</a>.) Set `X-Snowflake-Authorization-Token-Type` to `OAUTH` and set the Token to the auth token received from the external Auth server.

Set `X-Snowflake-Authorization-Token-Type` to `KEYPAIR_JWT` if the token is a key-pair authentication JWT.

Set `X-Snowflake-Authorization-Token-Type` to `OAUTH` if the token is snowflakeOAuth

Snowflake, Inc.

OpenAPI 3.0 specification for the Cortex REST API

Cortex Inference API

Perform LLM text completion inference, similar to snowflake.cortex.Complete.

Returns the LLMs available for the current session

Bad Request. The request payload is invalid or malformed. This happens if the application didn't send the correct request payload. The response body may include the error code and message indicating the actual cause. The application must reconstruct the request body for retry.

Unauthorized. The request is not authorized. This happens if the attached access token is invalid or missing. The response body may include the error code and message indicating the actual cause, e.g., expired, invalid token. The application must obtain a new access token for retry.

Forbidden. The request is forbidden. This can also happen if the request is made even if the API is not enabled.

Not Found. The request endpoint is not valid. This happens if the API endpoint does not exist, or if the API is not enabled.

Method Not Allowed. The request method doesn't match the supported API. This happens, for example, if the application calls the API with GET method but the endpoint accepts only POST.

Internal Server Error. The server hit an unrecoverable system error. The response body may include the error code and message for further guidance. The application owner may need to reach out the customer support.

Service Unavailable. The request was not processed due to server side timeouts. The application may retry with backoff. The jittered backoff is recommended.

Gateway Timeout. The request was not processed due to server side timeouts. The application may retry with backoff. The jittered backoff is recommended.

Cortex Inference

Returns the LLMs available for the current session¶

Resposta¶

Esquema do corpo de resposta¶

Perform LLM text completion inference¶

For more information

Resposta¶

Esquema do corpo de resposta¶