Foundation Models Text Generation API, gRPC: TextGenerationService.Completion

Статья создана

Yandex Cloud

Улучшена

Tania L.

Обновлена 21 февраля 2025 г.

gRPC request
CompletionRequest
CompletionOptions
ReasoningOptions
Message
ToolCallList
ToolCall
FunctionCall
ToolResultList
ToolResult
FunctionResult
Tool
FunctionTool
JsonSchema
CompletionResponse
Alternative
Message
ToolCallList
ToolCall
FunctionCall
ToolResultList
ToolResult
FunctionResult
ContentUsage
CompletionTokensDetails

A method for generating text completions in synchronous mode.

gRPC request

rpc Completion (CompletionRequest) returns (stream CompletionResponse)

CompletionRequest

{
  "model_uri": "string",
  "completion_options": {
    "stream": "bool",
    "temperature": "google.protobuf.DoubleValue",
    "max_tokens": "google.protobuf.Int64Value",
    "reasoning_options": {
      "mode": "ReasoningMode"
    }
  },
  "messages": [
    {
      "role": "string",
      // Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`
      "text": "string",
      "tool_call_list": {
        "tool_calls": [
          {
            // Includes only one of the fields `function_call`
            "function_call": {
              "name": "string",
              "arguments": "google.protobuf.Struct"
            }
            // end of the list of possible fields
          }
        ]
      },
      "tool_result_list": {
        "tool_results": [
          {
            // Includes only one of the fields `function_result`
            "function_result": {
              "name": "string",
              // Includes only one of the fields `content`
              "content": "string"
              // end of the list of possible fields
            }
            // end of the list of possible fields
          }
        ]
      }
      // end of the list of possible fields
    }
  ],
  "tools": [
    {
      // Includes only one of the fields `function`
      "function": {
        "name": "string",
        "description": "string",
        "parameters": "google.protobuf.Struct"
      }
      // end of the list of possible fields
    }
  ],
  // Includes only one of the fields `json_object`, `json_schema`
  "json_object": "bool",
  "json_schema": {
    "schema": "google.protobuf.Struct"
  }
  // end of the list of possible fields
}

Request for the service to generate text completion.

Field	Description
model_uri	string The ID of the model to be used for completion generation.
completion_options	CompletionOptions Configuration options for completion generation.
messages[]	Message A list of messages representing the context for the completion model.
tools[]	Tool List of tools that are available for the model to invoke during the completion generation. Note: This parameter is not yet supported and will be ignored if provided.
json_object	bool When set to true, the model will respond with a valid JSON object. Be sure to explicitly ask the model for JSON. Otherwise, it may generate excessive whitespace and run indefinitely until it reaches the token limit. Includes only one of the fields `json_object`, `json_schema`. Specifies the format of the model's response.
json_schema	JsonSchema Enforces a specific JSON structure for the model's response based on a provided schema. Includes only one of the fields `json_object`, `json_schema`. Specifies the format of the model's response.

CompletionOptions

Defines the options for completion generation.

Field	Description
stream	bool Enables streaming of partially generated text.
temperature	google.protobuf.DoubleValue Affects creativity and randomness of responses. Should be a double number between 0 (inclusive) and 1 (inclusive). Lower values produce more straightforward responses while higher values lead to increased creativity and randomness. Default temperature: 0.3
max_tokens	google.protobuf.Int64Value The limit on the number of tokens used for single completion generation. Must be greater than zero. This maximum allowed parameter value may depend on the model being used.
reasoning_options	ReasoningOptions Configures reasoning capabilities for the model, allowing it to perform internal reasoning before responding.

ReasoningOptions

Represents reasoning options that enable the model's ability to perform internal reasoning before generating a response.

Field

Description

mode

enum ReasoningMode

Specifies the reasoning mode to be used.

REASONING_MODE_UNSPECIFIED: Unspecified reasoning mode.
DISABLED: Disables reasoning. The model will generate a response without performing any internal reasoning.
ENABLED_HIDDEN: Enables reasoning in a hidden manner without exposing the reasoning steps to the user.

Message

A message object representing a wrapper over the inputs and outputs of the completion model.

Field	Description
role	string The ID of the message sender. Supported roles: `system`: Special role used to define the behaviour of the completion model. `assistant`: A role used by the model to generate responses. `user`: A role used by the user to describe requests to the model.
text	string Textual content of the message. Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`. Message content.
tool_call_list	ToolCallList List of tool calls made by the model as part of the response generation. Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`. Message content.
tool_result_list	ToolResultList List of tool results returned from external tools that were invoked by the model. Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`. Message content.

ToolCallList

Represents a list of tool calls.

Field

Description

tool_calls[]

ToolCall

A list of tool calls to be executed.

ToolCall

Represents a call to a tool.

Field

Description

function_call

FunctionCall

Represents a call to a function.

Includes only one of the fields function_call.

FunctionCall

Represents the invocation of a function with specific arguments.

Field

Description

name

string

The name of the function being called.

arguments

google.protobuf.Struct

The structured arguments passed to the function.
These arguments must adhere to the JSON Schema defined in the corresponding function's parameters.

ToolResultList

Represents a list of tool results.

Field

Description

tool_results[]

ToolResult

A list of tool results.

ToolResult

Represents the result of a tool call.

Field

Description

function_result

FunctionResult

Represents the result of a function call.

Includes only one of the fields function_result.

FunctionResult

Represents the result of a function call.

Field

Description

name

string

The name of the function that was executed.

content

string

The result of the function call, represented as a string.
This field can be used to store the output of the function.

Includes only one of the fields content.

Tool

Represents a tool that can be invoked during completion generation.

Field

Description

function

FunctionTool

Represents a function that can be called.

Includes only one of the fields function.

FunctionTool

Represents a function tool that can be invoked during completion generation.

Field	Description
name	string The name of the function.
description	string A description of the function's purpose or behavior.
parameters	google.protobuf.Struct A JSON Schema that defines the expected parameters for the function. The schema should describe the required fields, their types, and any constraints or default values.

JsonSchema

Represents the expected structure of the model's response using a JSON Schema.

Field

Description

schema

google.protobuf.Struct

The JSON Schema that the model's output must conform to.

CompletionResponse

{
  "alternatives": [
    {
      "message": {
        "role": "string",
        // Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`
        "text": "string",
        "tool_call_list": {
          "tool_calls": [
            {
              // Includes only one of the fields `function_call`
              "function_call": {
                "name": "string",
                "arguments": "google.protobuf.Struct"
              }
              // end of the list of possible fields
            }
          ]
        },
        "tool_result_list": {
          "tool_results": [
            {
              // Includes only one of the fields `function_result`
              "function_result": {
                "name": "string",
                // Includes only one of the fields `content`
                "content": "string"
                // end of the list of possible fields
              }
              // end of the list of possible fields
            }
          ]
        }
        // end of the list of possible fields
      },
      "status": "AlternativeStatus"
    }
  ],
  "usage": {
    "input_text_tokens": "int64",
    "completion_tokens": "int64",
    "total_tokens": "int64",
    "completion_tokens_details": {
      "reasoning_tokens": "int64"
    }
  },
  "model_version": "string"
}

Response containing generated text completions.

Field	Description
alternatives[]	Alternative A list of generated completion alternatives.
usage	ContentUsage A set of statistics describing the number of content tokens used by the completion model.
model_version	string The model version changes with each new releases.

Alternative

Represents a generated completion alternative, including its content and generation status.

Field

Description

message

Message

A message with the content of the alternative.

status

enum AlternativeStatus

The generation status of the alternative.

ALTERNATIVE_STATUS_UNSPECIFIED: Unspecified generation status.
ALTERNATIVE_STATUS_PARTIAL: Partially generated alternative.
ALTERNATIVE_STATUS_TRUNCATED_FINAL: Incomplete final alternative resulting from reaching the maximum allowed number of tokens.
ALTERNATIVE_STATUS_FINAL: Final alternative generated without running into any limits.
ALTERNATIVE_STATUS_CONTENT_FILTER: Generation was stopped due to the discovery of potentially sensitive content in the prompt or generated response.
To fix, modify the prompt and restart generation.
ALTERNATIVE_STATUS_TOOL_CALLS: Tools were invoked during the completion generation.

Message

A message object representing a wrapper over the inputs and outputs of the completion model.

Field	Description
role	string The ID of the message sender. Supported roles: `system`: Special role used to define the behaviour of the completion model. `assistant`: A role used by the model to generate responses. `user`: A role used by the user to describe requests to the model.
text	string Textual content of the message. Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`. Message content.
tool_call_list	ToolCallList List of tool calls made by the model as part of the response generation. Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`. Message content.
tool_result_list	ToolResultList List of tool results returned from external tools that were invoked by the model. Includes only one of the fields `text`, `tool_call_list`, `tool_result_list`. Message content.

ToolCallList

Represents a list of tool calls.

Field

Description

tool_calls[]

ToolCall

A list of tool calls to be executed.

ToolCall

Represents a call to a tool.

Field

Description

function_call

FunctionCall

Represents a call to a function.

Includes only one of the fields function_call.

FunctionCall

Represents the invocation of a function with specific arguments.

Field

Description

name

string

The name of the function being called.

arguments

google.protobuf.Struct

The structured arguments passed to the function.
These arguments must adhere to the JSON Schema defined in the corresponding function's parameters.

ToolResultList

Represents a list of tool results.

Field

Description

tool_results[]

ToolResult

A list of tool results.

ToolResult

Represents the result of a tool call.

Field

Description

function_result

FunctionResult

Represents the result of a function call.

Includes only one of the fields function_result.

FunctionResult

Represents the result of a function call.

Field

Description

name

string

The name of the function that was executed.

content

string

The result of the function call, represented as a string.
This field can be used to store the output of the function.

Includes only one of the fields content.

ContentUsage

An object representing the number of content tokens used by the completion model.

Field	Description
input_text_tokens	int64 The number of tokens in the textual part of the model input.
completion_tokens	int64 The number of tokens in the generated completion.
total_tokens	int64 The total number of tokens, including all input tokens and all generated tokens.
completion_tokens_details	CompletionTokensDetails Provides additional information about how the completion tokens were utilized.

CompletionTokensDetails

Provides additional information about how the completion tokens were utilized.

Field

Description

reasoning_tokens

int64

The number of tokens used specifically for internal reasoning performed by the model.

Foundation Models Text Generation API, gRPC: TextGenerationService.Completion

gRPC requestgRPC request

CompletionRequestCompletionRequest

CompletionOptionsCompletionOptions

ReasoningOptionsReasoningOptions

MessageMessage

ToolCallListToolCallList

ToolCallToolCall

FunctionCallFunctionCall

ToolResultListToolResultList

ToolResultToolResult

FunctionResultFunctionResult

ToolTool

FunctionToolFunctionTool

JsonSchemaJsonSchema

CompletionResponseCompletionResponse

AlternativeAlternative

MessageMessage

ToolCallListToolCallList

ToolCallToolCall

FunctionCallFunctionCall

ToolResultListToolResultList

ToolResultToolResult

FunctionResultFunctionResult

ContentUsageContentUsage

CompletionTokensDetailsCompletionTokensDetails

Была ли статья полезна?

gRPC request

CompletionRequest

CompletionOptions

ReasoningOptions

Message

ToolCallList

ToolCall

FunctionCall

ToolResultList

ToolResult

FunctionResult

Tool

FunctionTool

JsonSchema

CompletionResponse

Alternative

Message

ToolCallList

ToolCall

FunctionCall

ToolResultList

ToolResult

FunctionResult

ContentUsage

CompletionTokensDetails