Skip to content
OATSCHEMA

Docs / Schemas / llm-output / 0.7.0

LLM Output 0.7.0

Store a single interaction with an LLM (a Prompt/Response pair). Captures the input prompt, the resulting output, and the configuration parameters used at that moment.

This is an older version of llm-output.

The latest version is 0.8.0.

$id
https://oatschema.org/0.7.0/llm-output
Latest
https://oatschema.org/llm-output
Versions
Source
v0.7.0 on GitHub

Schema

llm-output.json Raw 
{    "x-license": {        "id": "Apache-2.0",        "notice": "Copyright 2025-2026 Dorsal Hub LTD",        "url": "https://github.com/oatschema/oats/blob/main/LICENSE"    },    "$schema": "https://json-schema.org/draft/2020-12/schema",    "$id": "https://oatschema.org/0.7.0/llm-output",    "title": "LLM Output",    "version": "0.7.0",    "description": "Store a single interaction with an LLM (a Prompt/Response pair). Captures the input prompt, the resulting output, and the configuration parameters used at that moment.",    "type": "object",    "properties": {        "model": {            "type": "string",            "description": "The ID or name of the model used (e.g., 'gpt-4o', 'claude-3-sonnet').",            "maxLength": 1024        },        "prompt": {            "type": "string",            "description": "The text input provided to the model. For contexts exceeding the character limit, this field should be truncated or a reference to the full storage location.",            "maxLength": 262144        },        "response_data": {            "description": "The output from the model. Non-plaintext output (e.g. JSON) should be serialized. For contexts exceeding the character limit, this field should be truncated or a reference to the full storage location.",            "type": "string",            "maxLength": 524288        },        "language": {            "type": "string",            "description": "The 3-letter ISO-639-3 language code of the response (e.g., 'eng', 'fra').",            "pattern": "^[a-z]{3}$",            "maxLength": 3        },        "score": {            "type": "number",            "description": "A normalized metric evaluating the response (-1.0 to 1.0). Positive values indicate high quality, acceptance, or positive sentiment. Negative values indicate rejection, low quality, or negative sentiment.",            "minimum": -1,            "maximum": 1        },        "score_explanation": {            "type": "string",            "description": "Defines the meaning and range of the 'score' field (e.g. 'Reward Model Score' or 'Confidence').",            "maxLength": 256        },        "generation_params": {            "type": "object",            "description": "Optional parameters sent in the API request to control the generation.",            "properties": {                "system_prompt": {                    "type": "string",                    "description": "An optional system-level instruction that defines the model's persona, style, or constraints. For contexts exceeding the character limit, this field should be truncated or a reference to the full storage location.",                    "maxLength": 262144                },                "temperature": {                    "type": "number",                    "description": "Controls randomness (e.g., 0.0 to 2.0).",                    "minimum": 0,                    "maximum": 2                },                "top_p": {                    "type": "number",                    "description": "Controls nucleus sampling (e.g., 0.0 to 1.0).",                    "minimum": 0,                    "maximum": 1                },                "top_k": {                    "type": "integer",                    "description": "Limits sampling to the k most likely tokens.",                    "minimum": 1                },                "max_tokens": {                    "type": "integer",                    "description": "The maximum number of tokens that were allowed to be generated.",                    "minimum": 1                },                "seed": {                    "type": "integer",                    "description": "The seed used for sampling for reproducible outputs."                },                "stop": {                    "description": "One or more sequences where the API will stop generating further tokens.",                    "anyOf": [                        {                            "type": "string",                            "maxLength": 128                        },                        {                            "type": "array",                            "items": {                                "type": "string",                                "maxLength": 128                            },                            "maxItems": 16                        }                    ]                },                "presence_penalty": {                    "type": "number",                    "description": "Penalty for new tokens based on whether they appear in the text so far (e.g., -2.0 to 2.0).",                    "minimum": -2,                    "maximum": 2                },                "frequency_penalty": {                    "type": "number",                    "description": "Penalty for new tokens based on their existing frequency in the text so far (e.g., -2.0 to 2.0).",                    "minimum": -2,                    "maximum": 2                },                "response_format": {                    "type": "object",                    "description": "Specifies the output format (e.g., 'text', 'json_object' or 'json_schema').",                    "properties": {                        "type": {                            "type": "string",                            "description": "The type of output format.",                            "enum": [                                "text",                                "json_object",                                "json_schema"                            ]                        }                    },                    "required": [                        "type"                    ],                    "additionalProperties": false                }            },            "additionalProperties": false        },        "generation_metadata": {            "type": "object",            "description": "Optional metadata returned by the API response, used for provenance and logging.",            "properties": {                "response_id": {                    "type": "string",                    "description": "The unique identifier for the generation response from the API provider.",                    "maxLength": 128                },                "created": {                    "type": "string",                    "format": "date-time",                    "maxLength": 128,                    "description": "The ISO-8601 timestamp of when the response was created."                },                "finish_reason": {                    "type": "string",                    "description": "The reason the model stopped generating tokens (e.g., 'stop', 'length', 'tool_calls').",                    "maxLength": 128                },                "system_fingerprint": {                    "type": "string",                    "description": "An identifier for the backend configuration that served the request, for reproducibility.",                    "maxLength": 128                },                "usage": {                    "type": "object",                    "description": "Token usage statistics for the generation request.",                    "properties": {                        "prompt_tokens": {                            "type": "integer",                            "description": "The number of tokens in the input.",                            "minimum": 0                        },                        "completion_tokens": {                            "type": "integer",                            "description": "The number of tokens generated.",                            "minimum": 0                        },                        "total_tokens": {                            "type": "integer",                            "description": "The total number of tokens used.",                            "minimum": 0                        },                        "cached_prompt_tokens": {                            "type": "integer",                            "description": "The number of input tokens served from a prompt cache.",                            "minimum": 0                        }                    },                    "required": [                        "prompt_tokens",                        "completion_tokens",                        "total_tokens"                    ],                    "additionalProperties": false                }            },            "additionalProperties": false        },        "attributes": {            "type": "object",            "description": "Arbitrary metadata relevant to this item.",            "maxProperties": 16,            "propertyNames": {                "maxLength": 128            },            "additionalProperties": {                "anyOf": [                    {                        "type": "string",                        "maxLength": 1024                    },                    {                        "type": "number"                    },                    {                        "type": "boolean"                    },                    {                        "type": "null"                    }                ]            }        }    },    "required": [        "model",        "response_data"    ],    "additionalProperties": false}

Examples

Records that pass validation against this schema.

llm-output-1
llm-output-1.json 
{  "model": "gpt-4o",  "prompt": "Extract the invoice number, total amount, and due date from this document.",  "response_data": "{\"invoice_id\": \"INV-9528\", \"total_due\": 1450.75, \"due_date\": \"2025-10-31\"}",  "language": "eng",  "score": 0.95,  "generation_params": {    "system_prompt": "You are an expert financial document parser. You must only output a valid JSON object.",    "temperature": 0.0,    "max_tokens": 1000,    "response_format": {      "type": "json_object"    }  },  "generation_metadata": {    "response_id": "chatcmpl-9qA8ZypD4YcW1bF5c6e7g8H9iJkLmN",    "created": "2025-09-17T11:45:00Z",    "finish_reason": "stop",    "system_fingerprint": "fp_a24b9d3f8e",    "usage": {      "prompt_tokens": 820,      "completion_tokens": 45,      "total_tokens": 865    }  }}

Fields

  • model string required

    The ID or name of the model used (e.g., 'gpt-4o', 'claude-3-sonnet').

    • at most 1024 characters
  • prompt string

    The text input provided to the model. For contexts exceeding the character limit, this field should be truncated or a reference to the full storage location.

    • at most 262144 characters
  • response_data string required

    The output from the model. Non-plaintext output (e.g. JSON) should be serialized. For contexts exceeding the character limit, this field should be truncated or a reference to the full storage location.

    • at most 524288 characters
  • language string

    The 3-letter ISO-639-3 language code of the response (e.g., 'eng', 'fra').

    • at most 3 characters
    • matches ^[a-z]{3}$
  • score number

    A normalized metric evaluating the response (-1.0 to 1.0). Positive values indicate high quality, acceptance, or positive sentiment. Negative values indicate rejection, low quality, or negative sentiment.

    • -1 to 1
  • score_explanation string

    Defines the meaning and range of the 'score' field (e.g. 'Reward Model Score' or 'Confidence').

    • at most 256 characters
  • generation_params object

    Optional parameters sent in the API request to control the generation.

    • system_prompt string

      An optional system-level instruction that defines the model's persona, style, or constraints. For contexts exceeding the character limit, this field should be truncated or a reference to the full storage location.

      • at most 262144 characters
    • temperature number

      Controls randomness (e.g., 0.0 to 2.0).

      • 0 to 2
    • top_p number

      Controls nucleus sampling (e.g., 0.0 to 1.0).

      • 0 to 1
    • top_k integer

      Limits sampling to the k most likely tokens.

      • ≥ 1
    • max_tokens integer

      The maximum number of tokens that were allowed to be generated.

      • ≥ 1
    • seed integer

      The seed used for sampling for reproducible outputs.

    • stop string | array of string

      One or more sequences where the API will stop generating further tokens.

      • string: at most 128 characters
      • array of string: at most 16 items
      • array of string: each at most 128 characters
    • presence_penalty number

      Penalty for new tokens based on whether they appear in the text so far (e.g., -2.0 to 2.0).

      • -2 to 2
    • frequency_penalty number

      Penalty for new tokens based on their existing frequency in the text so far (e.g., -2.0 to 2.0).

      • -2 to 2
    • response_format object

      Specifies the output format (e.g., 'text', 'json_object' or 'json_schema').

      • type string required

        The type of output format.

        Allowed values: "text", "json_object", "json_schema"

  • generation_metadata object

    Optional metadata returned by the API response, used for provenance and logging.

    • response_id string

      The unique identifier for the generation response from the API provider.

      • at most 128 characters
    • created string

      The ISO-8601 timestamp of when the response was created.

      • at most 128 characters
      • format date-time
    • finish_reason string

      The reason the model stopped generating tokens (e.g., 'stop', 'length', 'tool_calls').

      • at most 128 characters
    • system_fingerprint string

      An identifier for the backend configuration that served the request, for reproducibility.

      • at most 128 characters
    • usage object

      Token usage statistics for the generation request.

      • prompt_tokens integer required

        The number of tokens in the input.

        • ≥ 0
      • completion_tokens integer required

        The number of tokens generated.

        • ≥ 0
      • total_tokens integer required

        The total number of tokens used.

        • ≥ 0
      • cached_prompt_tokens integer

        The number of input tokens served from a prompt cache.

        • ≥ 0
  • attributes object

    Arbitrary metadata relevant to this item.

    • at most 16 properties
    • keys at most 128 characters
    • * string | number | boolean | null

      Any other key.

      • string: at most 1024 characters

Copyright 2025-2026 Dorsal Hub LTD. Licensed under Apache-2.0.