SaveLogsRequestSchema

creation_timestringrequired

UTC timestamp of the original inference in ISO 8601 format (e.g. 2025-06-05T15:50:00Z).

messages object[]required

The conversation turns to save as a training example. Each message must have role (system, user, or assistant) and content. The total token count for all messages must not exceed 8,192 tokens.

Possible values: >= 1

Array [

rolestringrequired

The role of the message author. system sets the assistant's behavior and persona. user represents input from the human turn. assistant represents a prior model response included for multi-turn context.

Possible values: [system, user, assistant]

contentstring

The text content of the message.

]

modelstringrequired

Name of the model that produced this inference output.

Example: Llama-3.3-70B-Instruct

kwargsobject

Additional metadata key-value pairs to store with the log entry.

Default: null

make_synthetic_versionboolean

When true, creates a synthetic data version of this log entry.

Default: null

tagsstring

Optional tags to attach to this log entry for later dataset filtering.

Default: null

usageobject

Token usage statistics for the logged inference call.

Default: null

SaveLogsRequestSchema

{
  "model": "Llama-3.3-70B-Instruct",
  "creation_time": "2025-01-15T10:30:00.000000+00:00",
  "messages": [
    {
      "role": "user",
      "content": "What is the capital of France?"
    }
  ],
  "tags": "apitest"
}